Thu, 14 Apr 2011 13:45:41 -0700
Merge
1 //
2 // Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 //
5 // This code is free software; you can redistribute it and/or modify it
6 // under the terms of the GNU General Public License version 2 only, as
7 // published by the Free Software Foundation.
8 //
9 // This code is distributed in the hope that it will be useful, but WITHOUT
10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 // version 2 for more details (a copy is included in the LICENSE file that
13 // accompanied this code).
14 //
15 // You should have received a copy of the GNU General Public License version
16 // 2 along with this work; if not, write to the Free Software Foundation,
17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 //
19 // Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 // or visit www.oracle.com if you need additional information or have any
21 // questions.
22 //
23 //
25 // AMD64 Architecture Description File
27 //----------REGISTER DEFINITION BLOCK------------------------------------------
28 // This information is used by the matcher and the register allocator to
29 // describe individual registers and classes of registers within the target
30 // archtecture.
32 register %{
33 //----------Architecture Description Register Definitions----------------------
34 // General Registers
35 // "reg_def" name ( register save type, C convention save type,
36 // ideal register type, encoding );
37 // Register Save Types:
38 //
39 // NS = No-Save: The register allocator assumes that these registers
40 // can be used without saving upon entry to the method, &
41 // that they do not need to be saved at call sites.
42 //
43 // SOC = Save-On-Call: The register allocator assumes that these registers
44 // can be used without saving upon entry to the method,
45 // but that they must be saved at call sites.
46 //
47 // SOE = Save-On-Entry: The register allocator assumes that these registers
48 // must be saved before using them upon entry to the
49 // method, but they do not need to be saved at call
50 // sites.
51 //
52 // AS = Always-Save: The register allocator assumes that these registers
53 // must be saved before using them upon entry to the
54 // method, & that they must be saved at call sites.
55 //
56 // Ideal Register Type is used to determine how to save & restore a
57 // register. Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
58 // spilled with LoadP/StoreP. If the register supports both, use Op_RegI.
59 //
60 // The encoding number is the actual bit-pattern placed into the opcodes.
62 // General Registers
63 // R8-R15 must be encoded with REX. (RSP, RBP, RSI, RDI need REX when
64 // used as byte registers)
66 // Previously set RBX, RSI, and RDI as save-on-entry for java code
67 // Turn off SOE in java-code due to frequent use of uncommon-traps.
68 // Now that allocator is better, turn on RSI and RDI as SOE registers.
70 reg_def RAX (SOC, SOC, Op_RegI, 0, rax->as_VMReg());
71 reg_def RAX_H(SOC, SOC, Op_RegI, 0, rax->as_VMReg()->next());
73 reg_def RCX (SOC, SOC, Op_RegI, 1, rcx->as_VMReg());
74 reg_def RCX_H(SOC, SOC, Op_RegI, 1, rcx->as_VMReg()->next());
76 reg_def RDX (SOC, SOC, Op_RegI, 2, rdx->as_VMReg());
77 reg_def RDX_H(SOC, SOC, Op_RegI, 2, rdx->as_VMReg()->next());
79 reg_def RBX (SOC, SOE, Op_RegI, 3, rbx->as_VMReg());
80 reg_def RBX_H(SOC, SOE, Op_RegI, 3, rbx->as_VMReg()->next());
82 reg_def RSP (NS, NS, Op_RegI, 4, rsp->as_VMReg());
83 reg_def RSP_H(NS, NS, Op_RegI, 4, rsp->as_VMReg()->next());
85 // now that adapter frames are gone RBP is always saved and restored by the prolog/epilog code
86 reg_def RBP (NS, SOE, Op_RegI, 5, rbp->as_VMReg());
87 reg_def RBP_H(NS, SOE, Op_RegI, 5, rbp->as_VMReg()->next());
89 #ifdef _WIN64
91 reg_def RSI (SOC, SOE, Op_RegI, 6, rsi->as_VMReg());
92 reg_def RSI_H(SOC, SOE, Op_RegI, 6, rsi->as_VMReg()->next());
94 reg_def RDI (SOC, SOE, Op_RegI, 7, rdi->as_VMReg());
95 reg_def RDI_H(SOC, SOE, Op_RegI, 7, rdi->as_VMReg()->next());
97 #else
99 reg_def RSI (SOC, SOC, Op_RegI, 6, rsi->as_VMReg());
100 reg_def RSI_H(SOC, SOC, Op_RegI, 6, rsi->as_VMReg()->next());
102 reg_def RDI (SOC, SOC, Op_RegI, 7, rdi->as_VMReg());
103 reg_def RDI_H(SOC, SOC, Op_RegI, 7, rdi->as_VMReg()->next());
105 #endif
107 reg_def R8 (SOC, SOC, Op_RegI, 8, r8->as_VMReg());
108 reg_def R8_H (SOC, SOC, Op_RegI, 8, r8->as_VMReg()->next());
110 reg_def R9 (SOC, SOC, Op_RegI, 9, r9->as_VMReg());
111 reg_def R9_H (SOC, SOC, Op_RegI, 9, r9->as_VMReg()->next());
113 reg_def R10 (SOC, SOC, Op_RegI, 10, r10->as_VMReg());
114 reg_def R10_H(SOC, SOC, Op_RegI, 10, r10->as_VMReg()->next());
116 reg_def R11 (SOC, SOC, Op_RegI, 11, r11->as_VMReg());
117 reg_def R11_H(SOC, SOC, Op_RegI, 11, r11->as_VMReg()->next());
119 reg_def R12 (SOC, SOE, Op_RegI, 12, r12->as_VMReg());
120 reg_def R12_H(SOC, SOE, Op_RegI, 12, r12->as_VMReg()->next());
122 reg_def R13 (SOC, SOE, Op_RegI, 13, r13->as_VMReg());
123 reg_def R13_H(SOC, SOE, Op_RegI, 13, r13->as_VMReg()->next());
125 reg_def R14 (SOC, SOE, Op_RegI, 14, r14->as_VMReg());
126 reg_def R14_H(SOC, SOE, Op_RegI, 14, r14->as_VMReg()->next());
128 reg_def R15 (SOC, SOE, Op_RegI, 15, r15->as_VMReg());
129 reg_def R15_H(SOC, SOE, Op_RegI, 15, r15->as_VMReg()->next());
132 // Floating Point Registers
134 // XMM registers. 128-bit registers or 4 words each, labeled (a)-d.
135 // Word a in each register holds a Float, words ab hold a Double. We
136 // currently do not use the SIMD capabilities, so registers cd are
137 // unused at the moment.
138 // XMM8-XMM15 must be encoded with REX.
139 // Linux ABI: No register preserved across function calls
140 // XMM0-XMM7 might hold parameters
141 // Windows ABI: XMM6-XMM15 preserved across function calls
142 // XMM0-XMM3 might hold parameters
144 reg_def XMM0 (SOC, SOC, Op_RegF, 0, xmm0->as_VMReg());
145 reg_def XMM0_H (SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next());
147 reg_def XMM1 (SOC, SOC, Op_RegF, 1, xmm1->as_VMReg());
148 reg_def XMM1_H (SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next());
150 reg_def XMM2 (SOC, SOC, Op_RegF, 2, xmm2->as_VMReg());
151 reg_def XMM2_H (SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next());
153 reg_def XMM3 (SOC, SOC, Op_RegF, 3, xmm3->as_VMReg());
154 reg_def XMM3_H (SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next());
156 reg_def XMM4 (SOC, SOC, Op_RegF, 4, xmm4->as_VMReg());
157 reg_def XMM4_H (SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next());
159 reg_def XMM5 (SOC, SOC, Op_RegF, 5, xmm5->as_VMReg());
160 reg_def XMM5_H (SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next());
162 #ifdef _WIN64
164 reg_def XMM6 (SOC, SOE, Op_RegF, 6, xmm6->as_VMReg());
165 reg_def XMM6_H (SOC, SOE, Op_RegF, 6, xmm6->as_VMReg()->next());
167 reg_def XMM7 (SOC, SOE, Op_RegF, 7, xmm7->as_VMReg());
168 reg_def XMM7_H (SOC, SOE, Op_RegF, 7, xmm7->as_VMReg()->next());
170 reg_def XMM8 (SOC, SOE, Op_RegF, 8, xmm8->as_VMReg());
171 reg_def XMM8_H (SOC, SOE, Op_RegF, 8, xmm8->as_VMReg()->next());
173 reg_def XMM9 (SOC, SOE, Op_RegF, 9, xmm9->as_VMReg());
174 reg_def XMM9_H (SOC, SOE, Op_RegF, 9, xmm9->as_VMReg()->next());
176 reg_def XMM10 (SOC, SOE, Op_RegF, 10, xmm10->as_VMReg());
177 reg_def XMM10_H(SOC, SOE, Op_RegF, 10, xmm10->as_VMReg()->next());
179 reg_def XMM11 (SOC, SOE, Op_RegF, 11, xmm11->as_VMReg());
180 reg_def XMM11_H(SOC, SOE, Op_RegF, 11, xmm11->as_VMReg()->next());
182 reg_def XMM12 (SOC, SOE, Op_RegF, 12, xmm12->as_VMReg());
183 reg_def XMM12_H(SOC, SOE, Op_RegF, 12, xmm12->as_VMReg()->next());
185 reg_def XMM13 (SOC, SOE, Op_RegF, 13, xmm13->as_VMReg());
186 reg_def XMM13_H(SOC, SOE, Op_RegF, 13, xmm13->as_VMReg()->next());
188 reg_def XMM14 (SOC, SOE, Op_RegF, 14, xmm14->as_VMReg());
189 reg_def XMM14_H(SOC, SOE, Op_RegF, 14, xmm14->as_VMReg()->next());
191 reg_def XMM15 (SOC, SOE, Op_RegF, 15, xmm15->as_VMReg());
192 reg_def XMM15_H(SOC, SOE, Op_RegF, 15, xmm15->as_VMReg()->next());
194 #else
196 reg_def XMM6 (SOC, SOC, Op_RegF, 6, xmm6->as_VMReg());
197 reg_def XMM6_H (SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next());
199 reg_def XMM7 (SOC, SOC, Op_RegF, 7, xmm7->as_VMReg());
200 reg_def XMM7_H (SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next());
202 reg_def XMM8 (SOC, SOC, Op_RegF, 8, xmm8->as_VMReg());
203 reg_def XMM8_H (SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next());
205 reg_def XMM9 (SOC, SOC, Op_RegF, 9, xmm9->as_VMReg());
206 reg_def XMM9_H (SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next());
208 reg_def XMM10 (SOC, SOC, Op_RegF, 10, xmm10->as_VMReg());
209 reg_def XMM10_H(SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next());
211 reg_def XMM11 (SOC, SOC, Op_RegF, 11, xmm11->as_VMReg());
212 reg_def XMM11_H(SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next());
214 reg_def XMM12 (SOC, SOC, Op_RegF, 12, xmm12->as_VMReg());
215 reg_def XMM12_H(SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next());
217 reg_def XMM13 (SOC, SOC, Op_RegF, 13, xmm13->as_VMReg());
218 reg_def XMM13_H(SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next());
220 reg_def XMM14 (SOC, SOC, Op_RegF, 14, xmm14->as_VMReg());
221 reg_def XMM14_H(SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next());
223 reg_def XMM15 (SOC, SOC, Op_RegF, 15, xmm15->as_VMReg());
224 reg_def XMM15_H(SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next());
226 #endif // _WIN64
228 reg_def RFLAGS(SOC, SOC, 0, 16, VMRegImpl::Bad());
230 // Specify priority of register selection within phases of register
231 // allocation. Highest priority is first. A useful heuristic is to
232 // give registers a low priority when they are required by machine
233 // instructions, like EAX and EDX on I486, and choose no-save registers
234 // before save-on-call, & save-on-call before save-on-entry. Registers
235 // which participate in fixed calling sequences should come last.
236 // Registers which are used as pairs must fall on an even boundary.
238 alloc_class chunk0(R10, R10_H,
239 R11, R11_H,
240 R8, R8_H,
241 R9, R9_H,
242 R12, R12_H,
243 RCX, RCX_H,
244 RBX, RBX_H,
245 RDI, RDI_H,
246 RDX, RDX_H,
247 RSI, RSI_H,
248 RAX, RAX_H,
249 RBP, RBP_H,
250 R13, R13_H,
251 R14, R14_H,
252 R15, R15_H,
253 RSP, RSP_H);
255 // XXX probably use 8-15 first on Linux
256 alloc_class chunk1(XMM0, XMM0_H,
257 XMM1, XMM1_H,
258 XMM2, XMM2_H,
259 XMM3, XMM3_H,
260 XMM4, XMM4_H,
261 XMM5, XMM5_H,
262 XMM6, XMM6_H,
263 XMM7, XMM7_H,
264 XMM8, XMM8_H,
265 XMM9, XMM9_H,
266 XMM10, XMM10_H,
267 XMM11, XMM11_H,
268 XMM12, XMM12_H,
269 XMM13, XMM13_H,
270 XMM14, XMM14_H,
271 XMM15, XMM15_H);
273 alloc_class chunk2(RFLAGS);
276 //----------Architecture Description Register Classes--------------------------
277 // Several register classes are automatically defined based upon information in
278 // this architecture description.
279 // 1) reg_class inline_cache_reg ( /* as def'd in frame section */ )
280 // 2) reg_class compiler_method_oop_reg ( /* as def'd in frame section */ )
281 // 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ )
282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
283 //
285 // Class for all pointer registers (including RSP)
286 reg_class any_reg(RAX, RAX_H,
287 RDX, RDX_H,
288 RBP, RBP_H,
289 RDI, RDI_H,
290 RSI, RSI_H,
291 RCX, RCX_H,
292 RBX, RBX_H,
293 RSP, RSP_H,
294 R8, R8_H,
295 R9, R9_H,
296 R10, R10_H,
297 R11, R11_H,
298 R12, R12_H,
299 R13, R13_H,
300 R14, R14_H,
301 R15, R15_H);
303 // Class for all pointer registers except RSP
304 reg_class ptr_reg(RAX, RAX_H,
305 RDX, RDX_H,
306 RBP, RBP_H,
307 RDI, RDI_H,
308 RSI, RSI_H,
309 RCX, RCX_H,
310 RBX, RBX_H,
311 R8, R8_H,
312 R9, R9_H,
313 R10, R10_H,
314 R11, R11_H,
315 R13, R13_H,
316 R14, R14_H);
318 // Class for all pointer registers except RAX and RSP
319 reg_class ptr_no_rax_reg(RDX, RDX_H,
320 RBP, RBP_H,
321 RDI, RDI_H,
322 RSI, RSI_H,
323 RCX, RCX_H,
324 RBX, RBX_H,
325 R8, R8_H,
326 R9, R9_H,
327 R10, R10_H,
328 R11, R11_H,
329 R13, R13_H,
330 R14, R14_H);
332 reg_class ptr_no_rbp_reg(RDX, RDX_H,
333 RAX, RAX_H,
334 RDI, RDI_H,
335 RSI, RSI_H,
336 RCX, RCX_H,
337 RBX, RBX_H,
338 R8, R8_H,
339 R9, R9_H,
340 R10, R10_H,
341 R11, R11_H,
342 R13, R13_H,
343 R14, R14_H);
345 // Class for all pointer registers except RAX, RBX and RSP
346 reg_class ptr_no_rax_rbx_reg(RDX, RDX_H,
347 RBP, RBP_H,
348 RDI, RDI_H,
349 RSI, RSI_H,
350 RCX, RCX_H,
351 R8, R8_H,
352 R9, R9_H,
353 R10, R10_H,
354 R11, R11_H,
355 R13, R13_H,
356 R14, R14_H);
358 // Singleton class for RAX pointer register
359 reg_class ptr_rax_reg(RAX, RAX_H);
361 // Singleton class for RBX pointer register
362 reg_class ptr_rbx_reg(RBX, RBX_H);
364 // Singleton class for RSI pointer register
365 reg_class ptr_rsi_reg(RSI, RSI_H);
367 // Singleton class for RDI pointer register
368 reg_class ptr_rdi_reg(RDI, RDI_H);
370 // Singleton class for RBP pointer register
371 reg_class ptr_rbp_reg(RBP, RBP_H);
373 // Singleton class for stack pointer
374 reg_class ptr_rsp_reg(RSP, RSP_H);
376 // Singleton class for TLS pointer
377 reg_class ptr_r15_reg(R15, R15_H);
379 // Class for all long registers (except RSP)
380 reg_class long_reg(RAX, RAX_H,
381 RDX, RDX_H,
382 RBP, RBP_H,
383 RDI, RDI_H,
384 RSI, RSI_H,
385 RCX, RCX_H,
386 RBX, RBX_H,
387 R8, R8_H,
388 R9, R9_H,
389 R10, R10_H,
390 R11, R11_H,
391 R13, R13_H,
392 R14, R14_H);
394 // Class for all long registers except RAX, RDX (and RSP)
395 reg_class long_no_rax_rdx_reg(RBP, RBP_H,
396 RDI, RDI_H,
397 RSI, RSI_H,
398 RCX, RCX_H,
399 RBX, RBX_H,
400 R8, R8_H,
401 R9, R9_H,
402 R10, R10_H,
403 R11, R11_H,
404 R13, R13_H,
405 R14, R14_H);
407 // Class for all long registers except RCX (and RSP)
408 reg_class long_no_rcx_reg(RBP, RBP_H,
409 RDI, RDI_H,
410 RSI, RSI_H,
411 RAX, RAX_H,
412 RDX, RDX_H,
413 RBX, RBX_H,
414 R8, R8_H,
415 R9, R9_H,
416 R10, R10_H,
417 R11, R11_H,
418 R13, R13_H,
419 R14, R14_H);
421 // Class for all long registers except RAX (and RSP)
422 reg_class long_no_rax_reg(RBP, RBP_H,
423 RDX, RDX_H,
424 RDI, RDI_H,
425 RSI, RSI_H,
426 RCX, RCX_H,
427 RBX, RBX_H,
428 R8, R8_H,
429 R9, R9_H,
430 R10, R10_H,
431 R11, R11_H,
432 R13, R13_H,
433 R14, R14_H);
435 // Singleton class for RAX long register
436 reg_class long_rax_reg(RAX, RAX_H);
438 // Singleton class for RCX long register
439 reg_class long_rcx_reg(RCX, RCX_H);
441 // Singleton class for RDX long register
442 reg_class long_rdx_reg(RDX, RDX_H);
444 // Class for all int registers (except RSP)
445 reg_class int_reg(RAX,
446 RDX,
447 RBP,
448 RDI,
449 RSI,
450 RCX,
451 RBX,
452 R8,
453 R9,
454 R10,
455 R11,
456 R13,
457 R14);
459 // Class for all int registers except RCX (and RSP)
460 reg_class int_no_rcx_reg(RAX,
461 RDX,
462 RBP,
463 RDI,
464 RSI,
465 RBX,
466 R8,
467 R9,
468 R10,
469 R11,
470 R13,
471 R14);
473 // Class for all int registers except RAX, RDX (and RSP)
474 reg_class int_no_rax_rdx_reg(RBP,
475 RDI,
476 RSI,
477 RCX,
478 RBX,
479 R8,
480 R9,
481 R10,
482 R11,
483 R13,
484 R14);
486 // Singleton class for RAX int register
487 reg_class int_rax_reg(RAX);
489 // Singleton class for RBX int register
490 reg_class int_rbx_reg(RBX);
492 // Singleton class for RCX int register
493 reg_class int_rcx_reg(RCX);
495 // Singleton class for RCX int register
496 reg_class int_rdx_reg(RDX);
498 // Singleton class for RCX int register
499 reg_class int_rdi_reg(RDI);
501 // Singleton class for instruction pointer
502 // reg_class ip_reg(RIP);
504 // Singleton class for condition codes
505 reg_class int_flags(RFLAGS);
507 // Class for all float registers
508 reg_class float_reg(XMM0,
509 XMM1,
510 XMM2,
511 XMM3,
512 XMM4,
513 XMM5,
514 XMM6,
515 XMM7,
516 XMM8,
517 XMM9,
518 XMM10,
519 XMM11,
520 XMM12,
521 XMM13,
522 XMM14,
523 XMM15);
525 // Class for all double registers
526 reg_class double_reg(XMM0, XMM0_H,
527 XMM1, XMM1_H,
528 XMM2, XMM2_H,
529 XMM3, XMM3_H,
530 XMM4, XMM4_H,
531 XMM5, XMM5_H,
532 XMM6, XMM6_H,
533 XMM7, XMM7_H,
534 XMM8, XMM8_H,
535 XMM9, XMM9_H,
536 XMM10, XMM10_H,
537 XMM11, XMM11_H,
538 XMM12, XMM12_H,
539 XMM13, XMM13_H,
540 XMM14, XMM14_H,
541 XMM15, XMM15_H);
542 %}
545 //----------SOURCE BLOCK-------------------------------------------------------
546 // This is a block of C++ code which provides values, functions, and
547 // definitions necessary in the rest of the architecture description
548 source %{
549 #define RELOC_IMM64 Assembler::imm_operand
550 #define RELOC_DISP32 Assembler::disp32_operand
552 #define __ _masm.
554 static int preserve_SP_size() {
555 return LP64_ONLY(1 +) 2; // [rex,] op, rm(reg/reg)
556 }
558 // !!!!! Special hack to get all types of calls to specify the byte offset
559 // from the start of the call to the point where the return address
560 // will point.
561 int MachCallStaticJavaNode::ret_addr_offset()
562 {
563 int offset = 5; // 5 bytes from start of call to where return address points
564 if (_method_handle_invoke)
565 offset += preserve_SP_size();
566 return offset;
567 }
569 int MachCallDynamicJavaNode::ret_addr_offset()
570 {
571 return 15; // 15 bytes from start of call to where return address points
572 }
574 // In os_cpu .ad file
575 // int MachCallRuntimeNode::ret_addr_offset()
577 // Indicate if the safepoint node needs the polling page as an input,
578 // it does if the polling page is more than disp32 away.
579 bool SafePointNode::needs_polling_address_input()
580 {
581 return Assembler::is_polling_page_far();
582 }
584 //
585 // Compute padding required for nodes which need alignment
586 //
588 // The address of the call instruction needs to be 4-byte aligned to
589 // ensure that it does not span a cache line so that it can be patched.
590 int CallStaticJavaDirectNode::compute_padding(int current_offset) const
591 {
592 current_offset += 1; // skip call opcode byte
593 return round_to(current_offset, alignment_required()) - current_offset;
594 }
596 // The address of the call instruction needs to be 4-byte aligned to
597 // ensure that it does not span a cache line so that it can be patched.
598 int CallStaticJavaHandleNode::compute_padding(int current_offset) const
599 {
600 current_offset += preserve_SP_size(); // skip mov rbp, rsp
601 current_offset += 1; // skip call opcode byte
602 return round_to(current_offset, alignment_required()) - current_offset;
603 }
605 // The address of the call instruction needs to be 4-byte aligned to
606 // ensure that it does not span a cache line so that it can be patched.
607 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
608 {
609 current_offset += 11; // skip movq instruction + call opcode byte
610 return round_to(current_offset, alignment_required()) - current_offset;
611 }
613 #ifndef PRODUCT
614 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const
615 {
616 st->print("INT3");
617 }
618 #endif
620 // EMIT_RM()
621 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3) {
622 unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3);
623 cbuf.insts()->emit_int8(c);
624 }
626 // EMIT_CC()
627 void emit_cc(CodeBuffer &cbuf, int f1, int f2) {
628 unsigned char c = (unsigned char) (f1 | f2);
629 cbuf.insts()->emit_int8(c);
630 }
632 // EMIT_OPCODE()
633 void emit_opcode(CodeBuffer &cbuf, int code) {
634 cbuf.insts()->emit_int8((unsigned char) code);
635 }
637 // EMIT_OPCODE() w/ relocation information
638 void emit_opcode(CodeBuffer &cbuf,
639 int code, relocInfo::relocType reloc, int offset, int format)
640 {
641 cbuf.relocate(cbuf.insts_mark() + offset, reloc, format);
642 emit_opcode(cbuf, code);
643 }
645 // EMIT_D8()
646 void emit_d8(CodeBuffer &cbuf, int d8) {
647 cbuf.insts()->emit_int8((unsigned char) d8);
648 }
650 // EMIT_D16()
651 void emit_d16(CodeBuffer &cbuf, int d16) {
652 cbuf.insts()->emit_int16(d16);
653 }
655 // EMIT_D32()
656 void emit_d32(CodeBuffer &cbuf, int d32) {
657 cbuf.insts()->emit_int32(d32);
658 }
660 // EMIT_D64()
661 void emit_d64(CodeBuffer &cbuf, int64_t d64) {
662 cbuf.insts()->emit_int64(d64);
663 }
665 // emit 32 bit value and construct relocation entry from relocInfo::relocType
666 void emit_d32_reloc(CodeBuffer& cbuf,
667 int d32,
668 relocInfo::relocType reloc,
669 int format)
670 {
671 assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc");
672 cbuf.relocate(cbuf.insts_mark(), reloc, format);
673 cbuf.insts()->emit_int32(d32);
674 }
676 // emit 32 bit value and construct relocation entry from RelocationHolder
677 void emit_d32_reloc(CodeBuffer& cbuf, int d32, RelocationHolder const& rspec, int format) {
678 #ifdef ASSERT
679 if (rspec.reloc()->type() == relocInfo::oop_type &&
680 d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) {
681 assert(oop((intptr_t)d32)->is_oop() && (ScavengeRootsInCode || !oop((intptr_t)d32)->is_scavengable()), "cannot embed scavengable oops in code");
682 }
683 #endif
684 cbuf.relocate(cbuf.insts_mark(), rspec, format);
685 cbuf.insts()->emit_int32(d32);
686 }
688 void emit_d32_reloc(CodeBuffer& cbuf, address addr) {
689 address next_ip = cbuf.insts_end() + 4;
690 emit_d32_reloc(cbuf, (int) (addr - next_ip),
691 external_word_Relocation::spec(addr),
692 RELOC_DISP32);
693 }
696 // emit 64 bit value and construct relocation entry from relocInfo::relocType
697 void emit_d64_reloc(CodeBuffer& cbuf, int64_t d64, relocInfo::relocType reloc, int format) {
698 cbuf.relocate(cbuf.insts_mark(), reloc, format);
699 cbuf.insts()->emit_int64(d64);
700 }
702 // emit 64 bit value and construct relocation entry from RelocationHolder
703 void emit_d64_reloc(CodeBuffer& cbuf, int64_t d64, RelocationHolder const& rspec, int format) {
704 #ifdef ASSERT
705 if (rspec.reloc()->type() == relocInfo::oop_type &&
706 d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) {
707 assert(oop(d64)->is_oop() && (ScavengeRootsInCode || !oop(d64)->is_scavengable()),
708 "cannot embed scavengable oops in code");
709 }
710 #endif
711 cbuf.relocate(cbuf.insts_mark(), rspec, format);
712 cbuf.insts()->emit_int64(d64);
713 }
715 // Access stack slot for load or store
716 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp)
717 {
718 emit_opcode(cbuf, opcode); // (e.g., FILD [RSP+src])
719 if (-0x80 <= disp && disp < 0x80) {
720 emit_rm(cbuf, 0x01, rm_field, RSP_enc); // R/M byte
721 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
722 emit_d8(cbuf, disp); // Displacement // R/M byte
723 } else {
724 emit_rm(cbuf, 0x02, rm_field, RSP_enc); // R/M byte
725 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
726 emit_d32(cbuf, disp); // Displacement // R/M byte
727 }
728 }
730 // rRegI ereg, memory mem) %{ // emit_reg_mem
731 void encode_RegMem(CodeBuffer &cbuf,
732 int reg,
733 int base, int index, int scale, int disp, bool disp_is_oop)
734 {
735 assert(!disp_is_oop, "cannot have disp");
736 int regenc = reg & 7;
737 int baseenc = base & 7;
738 int indexenc = index & 7;
740 // There is no index & no scale, use form without SIB byte
741 if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) {
742 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
743 if (disp == 0 && base != RBP_enc && base != R13_enc) {
744 emit_rm(cbuf, 0x0, regenc, baseenc); // *
745 } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
746 // If 8-bit displacement, mode 0x1
747 emit_rm(cbuf, 0x1, regenc, baseenc); // *
748 emit_d8(cbuf, disp);
749 } else {
750 // If 32-bit displacement
751 if (base == -1) { // Special flag for absolute address
752 emit_rm(cbuf, 0x0, regenc, 0x5); // *
753 if (disp_is_oop) {
754 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
755 } else {
756 emit_d32(cbuf, disp);
757 }
758 } else {
759 // Normal base + offset
760 emit_rm(cbuf, 0x2, regenc, baseenc); // *
761 if (disp_is_oop) {
762 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
763 } else {
764 emit_d32(cbuf, disp);
765 }
766 }
767 }
768 } else {
769 // Else, encode with the SIB byte
770 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
771 if (disp == 0 && base != RBP_enc && base != R13_enc) {
772 // If no displacement
773 emit_rm(cbuf, 0x0, regenc, 0x4); // *
774 emit_rm(cbuf, scale, indexenc, baseenc);
775 } else {
776 if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
777 // If 8-bit displacement, mode 0x1
778 emit_rm(cbuf, 0x1, regenc, 0x4); // *
779 emit_rm(cbuf, scale, indexenc, baseenc);
780 emit_d8(cbuf, disp);
781 } else {
782 // If 32-bit displacement
783 if (base == 0x04 ) {
784 emit_rm(cbuf, 0x2, regenc, 0x4);
785 emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid???
786 } else {
787 emit_rm(cbuf, 0x2, regenc, 0x4);
788 emit_rm(cbuf, scale, indexenc, baseenc); // *
789 }
790 if (disp_is_oop) {
791 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
792 } else {
793 emit_d32(cbuf, disp);
794 }
795 }
796 }
797 }
798 }
800 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc)
801 {
802 if (dstenc != srcenc) {
803 if (dstenc < 8) {
804 if (srcenc >= 8) {
805 emit_opcode(cbuf, Assembler::REX_B);
806 srcenc -= 8;
807 }
808 } else {
809 if (srcenc < 8) {
810 emit_opcode(cbuf, Assembler::REX_R);
811 } else {
812 emit_opcode(cbuf, Assembler::REX_RB);
813 srcenc -= 8;
814 }
815 dstenc -= 8;
816 }
818 emit_opcode(cbuf, 0x8B);
819 emit_rm(cbuf, 0x3, dstenc, srcenc);
820 }
821 }
823 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
824 if( dst_encoding == src_encoding ) {
825 // reg-reg copy, use an empty encoding
826 } else {
827 MacroAssembler _masm(&cbuf);
829 __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding));
830 }
831 }
834 //=============================================================================
835 const bool Matcher::constant_table_absolute_addressing = true;
836 const RegMask& MachConstantBaseNode::_out_RegMask = RegMask::Empty;
838 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
839 // Empty encoding
840 }
842 uint MachConstantBaseNode::size(PhaseRegAlloc* ra_) const {
843 return 0;
844 }
846 #ifndef PRODUCT
847 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
848 st->print("# MachConstantBaseNode (empty encoding)");
849 }
850 #endif
853 //=============================================================================
854 #ifndef PRODUCT
855 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const
856 {
857 Compile* C = ra_->C;
859 int framesize = C->frame_slots() << LogBytesPerInt;
860 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
861 // Remove wordSize for return adr already pushed
862 // and another for the RBP we are going to save
863 framesize -= 2*wordSize;
864 bool need_nop = true;
866 // Calls to C2R adapters often do not accept exceptional returns.
867 // We require that their callers must bang for them. But be
868 // careful, because some VM calls (such as call site linkage) can
869 // use several kilobytes of stack. But the stack safety zone should
870 // account for that. See bugs 4446381, 4468289, 4497237.
871 if (C->need_stack_bang(framesize)) {
872 st->print_cr("# stack bang"); st->print("\t");
873 need_nop = false;
874 }
875 st->print_cr("pushq rbp"); st->print("\t");
877 if (VerifyStackAtCalls) {
878 // Majik cookie to verify stack depth
879 st->print_cr("pushq 0xffffffffbadb100d"
880 "\t# Majik cookie for stack depth check");
881 st->print("\t");
882 framesize -= wordSize; // Remove 2 for cookie
883 need_nop = false;
884 }
886 if (framesize) {
887 st->print("subq rsp, #%d\t# Create frame", framesize);
888 if (framesize < 0x80 && need_nop) {
889 st->print("\n\tnop\t# nop for patch_verified_entry");
890 }
891 }
892 }
893 #endif
895 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
896 {
897 Compile* C = ra_->C;
899 // WARNING: Initial instruction MUST be 5 bytes or longer so that
900 // NativeJump::patch_verified_entry will be able to patch out the entry
901 // code safely. The fldcw is ok at 6 bytes, the push to verify stack
902 // depth is ok at 5 bytes, the frame allocation can be either 3 or
903 // 6 bytes. So if we don't do the fldcw or the push then we must
904 // use the 6 byte frame allocation even if we have no frame. :-(
905 // If method sets FPU control word do it now
907 int framesize = C->frame_slots() << LogBytesPerInt;
908 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
909 // Remove wordSize for return adr already pushed
910 // and another for the RBP we are going to save
911 framesize -= 2*wordSize;
912 bool need_nop = true;
914 // Calls to C2R adapters often do not accept exceptional returns.
915 // We require that their callers must bang for them. But be
916 // careful, because some VM calls (such as call site linkage) can
917 // use several kilobytes of stack. But the stack safety zone should
918 // account for that. See bugs 4446381, 4468289, 4497237.
919 if (C->need_stack_bang(framesize)) {
920 MacroAssembler masm(&cbuf);
921 masm.generate_stack_overflow_check(framesize);
922 need_nop = false;
923 }
925 // We always push rbp so that on return to interpreter rbp will be
926 // restored correctly and we can correct the stack.
927 emit_opcode(cbuf, 0x50 | RBP_enc);
929 if (VerifyStackAtCalls) {
930 // Majik cookie to verify stack depth
931 emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d
932 emit_d32(cbuf, 0xbadb100d);
933 framesize -= wordSize; // Remove 2 for cookie
934 need_nop = false;
935 }
937 if (framesize) {
938 emit_opcode(cbuf, Assembler::REX_W);
939 if (framesize < 0x80) {
940 emit_opcode(cbuf, 0x83); // sub SP,#framesize
941 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
942 emit_d8(cbuf, framesize);
943 if (need_nop) {
944 emit_opcode(cbuf, 0x90); // nop
945 }
946 } else {
947 emit_opcode(cbuf, 0x81); // sub SP,#framesize
948 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
949 emit_d32(cbuf, framesize);
950 }
951 }
953 C->set_frame_complete(cbuf.insts_size());
955 #ifdef ASSERT
956 if (VerifyStackAtCalls) {
957 Label L;
958 MacroAssembler masm(&cbuf);
959 masm.push(rax);
960 masm.mov(rax, rsp);
961 masm.andptr(rax, StackAlignmentInBytes-1);
962 masm.cmpptr(rax, StackAlignmentInBytes-wordSize);
963 masm.pop(rax);
964 masm.jcc(Assembler::equal, L);
965 masm.stop("Stack is not properly aligned!");
966 masm.bind(L);
967 }
968 #endif
969 }
971 uint MachPrologNode::size(PhaseRegAlloc* ra_) const
972 {
973 return MachNode::size(ra_); // too many variables; just compute it
974 // the hard way
975 }
977 int MachPrologNode::reloc() const
978 {
979 return 0; // a large enough number
980 }
982 //=============================================================================
983 #ifndef PRODUCT
984 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const
985 {
986 Compile* C = ra_->C;
987 int framesize = C->frame_slots() << LogBytesPerInt;
988 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
989 // Remove word for return adr already pushed
990 // and RBP
991 framesize -= 2*wordSize;
993 if (framesize) {
994 st->print_cr("addq rsp, %d\t# Destroy frame", framesize);
995 st->print("\t");
996 }
998 st->print_cr("popq rbp");
999 if (do_polling() && C->is_method_compilation()) {
1000 st->print("\t");
1001 if (Assembler::is_polling_page_far()) {
1002 st->print_cr("movq rscratch1, #polling_page_address\n\t"
1003 "testl rax, [rscratch1]\t"
1004 "# Safepoint: poll for GC");
1005 } else {
1006 st->print_cr("testl rax, [rip + #offset_to_poll_page]\t"
1007 "# Safepoint: poll for GC");
1008 }
1009 }
1010 }
1011 #endif
1013 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1014 {
1015 Compile* C = ra_->C;
1016 int framesize = C->frame_slots() << LogBytesPerInt;
1017 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1018 // Remove word for return adr already pushed
1019 // and RBP
1020 framesize -= 2*wordSize;
1022 // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
1024 if (framesize) {
1025 emit_opcode(cbuf, Assembler::REX_W);
1026 if (framesize < 0x80) {
1027 emit_opcode(cbuf, 0x83); // addq rsp, #framesize
1028 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1029 emit_d8(cbuf, framesize);
1030 } else {
1031 emit_opcode(cbuf, 0x81); // addq rsp, #framesize
1032 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1033 emit_d32(cbuf, framesize);
1034 }
1035 }
1037 // popq rbp
1038 emit_opcode(cbuf, 0x58 | RBP_enc);
1040 if (do_polling() && C->is_method_compilation()) {
1041 MacroAssembler _masm(&cbuf);
1042 AddressLiteral polling_page(os::get_polling_page(), relocInfo::poll_return_type);
1043 if (Assembler::is_polling_page_far()) {
1044 __ lea(rscratch1, polling_page);
1045 __ relocate(relocInfo::poll_return_type);
1046 __ testl(rax, Address(rscratch1, 0));
1047 } else {
1048 __ testl(rax, polling_page);
1049 }
1050 }
1051 }
1053 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const
1054 {
1055 return MachNode::size(ra_); // too many variables; just compute it
1056 // the hard way
1057 }
1059 int MachEpilogNode::reloc() const
1060 {
1061 return 2; // a large enough number
1062 }
1064 const Pipeline* MachEpilogNode::pipeline() const
1065 {
1066 return MachNode::pipeline_class();
1067 }
1069 int MachEpilogNode::safepoint_offset() const
1070 {
1071 return 0;
1072 }
1074 //=============================================================================
1076 enum RC {
1077 rc_bad,
1078 rc_int,
1079 rc_float,
1080 rc_stack
1081 };
1083 static enum RC rc_class(OptoReg::Name reg)
1084 {
1085 if( !OptoReg::is_valid(reg) ) return rc_bad;
1087 if (OptoReg::is_stack(reg)) return rc_stack;
1089 VMReg r = OptoReg::as_VMReg(reg);
1091 if (r->is_Register()) return rc_int;
1093 assert(r->is_XMMRegister(), "must be");
1094 return rc_float;
1095 }
1097 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf,
1098 PhaseRegAlloc* ra_,
1099 bool do_size,
1100 outputStream* st) const
1101 {
1103 // Get registers to move
1104 OptoReg::Name src_second = ra_->get_reg_second(in(1));
1105 OptoReg::Name src_first = ra_->get_reg_first(in(1));
1106 OptoReg::Name dst_second = ra_->get_reg_second(this);
1107 OptoReg::Name dst_first = ra_->get_reg_first(this);
1109 enum RC src_second_rc = rc_class(src_second);
1110 enum RC src_first_rc = rc_class(src_first);
1111 enum RC dst_second_rc = rc_class(dst_second);
1112 enum RC dst_first_rc = rc_class(dst_first);
1114 assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first),
1115 "must move at least 1 register" );
1117 if (src_first == dst_first && src_second == dst_second) {
1118 // Self copy, no move
1119 return 0;
1120 } else if (src_first_rc == rc_stack) {
1121 // mem ->
1122 if (dst_first_rc == rc_stack) {
1123 // mem -> mem
1124 assert(src_second != dst_first, "overlap");
1125 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1126 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1127 // 64-bit
1128 int src_offset = ra_->reg2offset(src_first);
1129 int dst_offset = ra_->reg2offset(dst_first);
1130 if (cbuf) {
1131 emit_opcode(*cbuf, 0xFF);
1132 encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false);
1134 emit_opcode(*cbuf, 0x8F);
1135 encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false);
1137 #ifndef PRODUCT
1138 } else if (!do_size) {
1139 st->print("pushq [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
1140 "popq [rsp + #%d]",
1141 src_offset,
1142 dst_offset);
1143 #endif
1144 }
1145 return
1146 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) +
1147 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4));
1148 } else {
1149 // 32-bit
1150 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1151 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1152 // No pushl/popl, so:
1153 int src_offset = ra_->reg2offset(src_first);
1154 int dst_offset = ra_->reg2offset(dst_first);
1155 if (cbuf) {
1156 emit_opcode(*cbuf, Assembler::REX_W);
1157 emit_opcode(*cbuf, 0x89);
1158 emit_opcode(*cbuf, 0x44);
1159 emit_opcode(*cbuf, 0x24);
1160 emit_opcode(*cbuf, 0xF8);
1162 emit_opcode(*cbuf, 0x8B);
1163 encode_RegMem(*cbuf,
1164 RAX_enc,
1165 RSP_enc, 0x4, 0, src_offset,
1166 false);
1168 emit_opcode(*cbuf, 0x89);
1169 encode_RegMem(*cbuf,
1170 RAX_enc,
1171 RSP_enc, 0x4, 0, dst_offset,
1172 false);
1174 emit_opcode(*cbuf, Assembler::REX_W);
1175 emit_opcode(*cbuf, 0x8B);
1176 emit_opcode(*cbuf, 0x44);
1177 emit_opcode(*cbuf, 0x24);
1178 emit_opcode(*cbuf, 0xF8);
1180 #ifndef PRODUCT
1181 } else if (!do_size) {
1182 st->print("movq [rsp - #8], rax\t# 32-bit mem-mem spill\n\t"
1183 "movl rax, [rsp + #%d]\n\t"
1184 "movl [rsp + #%d], rax\n\t"
1185 "movq rax, [rsp - #8]",
1186 src_offset,
1187 dst_offset);
1188 #endif
1189 }
1190 return
1191 5 + // movq
1192 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl
1193 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl
1194 5; // movq
1195 }
1196 } else if (dst_first_rc == rc_int) {
1197 // mem -> gpr
1198 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1199 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1200 // 64-bit
1201 int offset = ra_->reg2offset(src_first);
1202 if (cbuf) {
1203 if (Matcher::_regEncode[dst_first] < 8) {
1204 emit_opcode(*cbuf, Assembler::REX_W);
1205 } else {
1206 emit_opcode(*cbuf, Assembler::REX_WR);
1207 }
1208 emit_opcode(*cbuf, 0x8B);
1209 encode_RegMem(*cbuf,
1210 Matcher::_regEncode[dst_first],
1211 RSP_enc, 0x4, 0, offset,
1212 false);
1213 #ifndef PRODUCT
1214 } else if (!do_size) {
1215 st->print("movq %s, [rsp + #%d]\t# spill",
1216 Matcher::regName[dst_first],
1217 offset);
1218 #endif
1219 }
1220 return
1221 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1222 } else {
1223 // 32-bit
1224 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1225 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1226 int offset = ra_->reg2offset(src_first);
1227 if (cbuf) {
1228 if (Matcher::_regEncode[dst_first] >= 8) {
1229 emit_opcode(*cbuf, Assembler::REX_R);
1230 }
1231 emit_opcode(*cbuf, 0x8B);
1232 encode_RegMem(*cbuf,
1233 Matcher::_regEncode[dst_first],
1234 RSP_enc, 0x4, 0, offset,
1235 false);
1236 #ifndef PRODUCT
1237 } else if (!do_size) {
1238 st->print("movl %s, [rsp + #%d]\t# spill",
1239 Matcher::regName[dst_first],
1240 offset);
1241 #endif
1242 }
1243 return
1244 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1245 ((Matcher::_regEncode[dst_first] < 8)
1246 ? 3
1247 : 4); // REX
1248 }
1249 } else if (dst_first_rc == rc_float) {
1250 // mem-> xmm
1251 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1252 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1253 // 64-bit
1254 int offset = ra_->reg2offset(src_first);
1255 if (cbuf) {
1256 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
1257 if (Matcher::_regEncode[dst_first] >= 8) {
1258 emit_opcode(*cbuf, Assembler::REX_R);
1259 }
1260 emit_opcode(*cbuf, 0x0F);
1261 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
1262 encode_RegMem(*cbuf,
1263 Matcher::_regEncode[dst_first],
1264 RSP_enc, 0x4, 0, offset,
1265 false);
1266 #ifndef PRODUCT
1267 } else if (!do_size) {
1268 st->print("%s %s, [rsp + #%d]\t# spill",
1269 UseXmmLoadAndClearUpper ? "movsd " : "movlpd",
1270 Matcher::regName[dst_first],
1271 offset);
1272 #endif
1273 }
1274 return
1275 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1276 ((Matcher::_regEncode[dst_first] < 8)
1277 ? 5
1278 : 6); // REX
1279 } else {
1280 // 32-bit
1281 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1282 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1283 int offset = ra_->reg2offset(src_first);
1284 if (cbuf) {
1285 emit_opcode(*cbuf, 0xF3);
1286 if (Matcher::_regEncode[dst_first] >= 8) {
1287 emit_opcode(*cbuf, Assembler::REX_R);
1288 }
1289 emit_opcode(*cbuf, 0x0F);
1290 emit_opcode(*cbuf, 0x10);
1291 encode_RegMem(*cbuf,
1292 Matcher::_regEncode[dst_first],
1293 RSP_enc, 0x4, 0, offset,
1294 false);
1295 #ifndef PRODUCT
1296 } else if (!do_size) {
1297 st->print("movss %s, [rsp + #%d]\t# spill",
1298 Matcher::regName[dst_first],
1299 offset);
1300 #endif
1301 }
1302 return
1303 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1304 ((Matcher::_regEncode[dst_first] < 8)
1305 ? 5
1306 : 6); // REX
1307 }
1308 }
1309 } else if (src_first_rc == rc_int) {
1310 // gpr ->
1311 if (dst_first_rc == rc_stack) {
1312 // gpr -> mem
1313 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1314 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1315 // 64-bit
1316 int offset = ra_->reg2offset(dst_first);
1317 if (cbuf) {
1318 if (Matcher::_regEncode[src_first] < 8) {
1319 emit_opcode(*cbuf, Assembler::REX_W);
1320 } else {
1321 emit_opcode(*cbuf, Assembler::REX_WR);
1322 }
1323 emit_opcode(*cbuf, 0x89);
1324 encode_RegMem(*cbuf,
1325 Matcher::_regEncode[src_first],
1326 RSP_enc, 0x4, 0, offset,
1327 false);
1328 #ifndef PRODUCT
1329 } else if (!do_size) {
1330 st->print("movq [rsp + #%d], %s\t# spill",
1331 offset,
1332 Matcher::regName[src_first]);
1333 #endif
1334 }
1335 return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1336 } else {
1337 // 32-bit
1338 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1339 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1340 int offset = ra_->reg2offset(dst_first);
1341 if (cbuf) {
1342 if (Matcher::_regEncode[src_first] >= 8) {
1343 emit_opcode(*cbuf, Assembler::REX_R);
1344 }
1345 emit_opcode(*cbuf, 0x89);
1346 encode_RegMem(*cbuf,
1347 Matcher::_regEncode[src_first],
1348 RSP_enc, 0x4, 0, offset,
1349 false);
1350 #ifndef PRODUCT
1351 } else if (!do_size) {
1352 st->print("movl [rsp + #%d], %s\t# spill",
1353 offset,
1354 Matcher::regName[src_first]);
1355 #endif
1356 }
1357 return
1358 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1359 ((Matcher::_regEncode[src_first] < 8)
1360 ? 3
1361 : 4); // REX
1362 }
1363 } else if (dst_first_rc == rc_int) {
1364 // gpr -> gpr
1365 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1366 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1367 // 64-bit
1368 if (cbuf) {
1369 if (Matcher::_regEncode[dst_first] < 8) {
1370 if (Matcher::_regEncode[src_first] < 8) {
1371 emit_opcode(*cbuf, Assembler::REX_W);
1372 } else {
1373 emit_opcode(*cbuf, Assembler::REX_WB);
1374 }
1375 } else {
1376 if (Matcher::_regEncode[src_first] < 8) {
1377 emit_opcode(*cbuf, Assembler::REX_WR);
1378 } else {
1379 emit_opcode(*cbuf, Assembler::REX_WRB);
1380 }
1381 }
1382 emit_opcode(*cbuf, 0x8B);
1383 emit_rm(*cbuf, 0x3,
1384 Matcher::_regEncode[dst_first] & 7,
1385 Matcher::_regEncode[src_first] & 7);
1386 #ifndef PRODUCT
1387 } else if (!do_size) {
1388 st->print("movq %s, %s\t# spill",
1389 Matcher::regName[dst_first],
1390 Matcher::regName[src_first]);
1391 #endif
1392 }
1393 return 3; // REX
1394 } else {
1395 // 32-bit
1396 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1397 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1398 if (cbuf) {
1399 if (Matcher::_regEncode[dst_first] < 8) {
1400 if (Matcher::_regEncode[src_first] >= 8) {
1401 emit_opcode(*cbuf, Assembler::REX_B);
1402 }
1403 } else {
1404 if (Matcher::_regEncode[src_first] < 8) {
1405 emit_opcode(*cbuf, Assembler::REX_R);
1406 } else {
1407 emit_opcode(*cbuf, Assembler::REX_RB);
1408 }
1409 }
1410 emit_opcode(*cbuf, 0x8B);
1411 emit_rm(*cbuf, 0x3,
1412 Matcher::_regEncode[dst_first] & 7,
1413 Matcher::_regEncode[src_first] & 7);
1414 #ifndef PRODUCT
1415 } else if (!do_size) {
1416 st->print("movl %s, %s\t# spill",
1417 Matcher::regName[dst_first],
1418 Matcher::regName[src_first]);
1419 #endif
1420 }
1421 return
1422 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1423 ? 2
1424 : 3; // REX
1425 }
1426 } else if (dst_first_rc == rc_float) {
1427 // gpr -> xmm
1428 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1429 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1430 // 64-bit
1431 if (cbuf) {
1432 emit_opcode(*cbuf, 0x66);
1433 if (Matcher::_regEncode[dst_first] < 8) {
1434 if (Matcher::_regEncode[src_first] < 8) {
1435 emit_opcode(*cbuf, Assembler::REX_W);
1436 } else {
1437 emit_opcode(*cbuf, Assembler::REX_WB);
1438 }
1439 } else {
1440 if (Matcher::_regEncode[src_first] < 8) {
1441 emit_opcode(*cbuf, Assembler::REX_WR);
1442 } else {
1443 emit_opcode(*cbuf, Assembler::REX_WRB);
1444 }
1445 }
1446 emit_opcode(*cbuf, 0x0F);
1447 emit_opcode(*cbuf, 0x6E);
1448 emit_rm(*cbuf, 0x3,
1449 Matcher::_regEncode[dst_first] & 7,
1450 Matcher::_regEncode[src_first] & 7);
1451 #ifndef PRODUCT
1452 } else if (!do_size) {
1453 st->print("movdq %s, %s\t# spill",
1454 Matcher::regName[dst_first],
1455 Matcher::regName[src_first]);
1456 #endif
1457 }
1458 return 5; // REX
1459 } else {
1460 // 32-bit
1461 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1462 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1463 if (cbuf) {
1464 emit_opcode(*cbuf, 0x66);
1465 if (Matcher::_regEncode[dst_first] < 8) {
1466 if (Matcher::_regEncode[src_first] >= 8) {
1467 emit_opcode(*cbuf, Assembler::REX_B);
1468 }
1469 } else {
1470 if (Matcher::_regEncode[src_first] < 8) {
1471 emit_opcode(*cbuf, Assembler::REX_R);
1472 } else {
1473 emit_opcode(*cbuf, Assembler::REX_RB);
1474 }
1475 }
1476 emit_opcode(*cbuf, 0x0F);
1477 emit_opcode(*cbuf, 0x6E);
1478 emit_rm(*cbuf, 0x3,
1479 Matcher::_regEncode[dst_first] & 7,
1480 Matcher::_regEncode[src_first] & 7);
1481 #ifndef PRODUCT
1482 } else if (!do_size) {
1483 st->print("movdl %s, %s\t# spill",
1484 Matcher::regName[dst_first],
1485 Matcher::regName[src_first]);
1486 #endif
1487 }
1488 return
1489 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1490 ? 4
1491 : 5; // REX
1492 }
1493 }
1494 } else if (src_first_rc == rc_float) {
1495 // xmm ->
1496 if (dst_first_rc == rc_stack) {
1497 // xmm -> mem
1498 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1499 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1500 // 64-bit
1501 int offset = ra_->reg2offset(dst_first);
1502 if (cbuf) {
1503 emit_opcode(*cbuf, 0xF2);
1504 if (Matcher::_regEncode[src_first] >= 8) {
1505 emit_opcode(*cbuf, Assembler::REX_R);
1506 }
1507 emit_opcode(*cbuf, 0x0F);
1508 emit_opcode(*cbuf, 0x11);
1509 encode_RegMem(*cbuf,
1510 Matcher::_regEncode[src_first],
1511 RSP_enc, 0x4, 0, offset,
1512 false);
1513 #ifndef PRODUCT
1514 } else if (!do_size) {
1515 st->print("movsd [rsp + #%d], %s\t# spill",
1516 offset,
1517 Matcher::regName[src_first]);
1518 #endif
1519 }
1520 return
1521 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1522 ((Matcher::_regEncode[src_first] < 8)
1523 ? 5
1524 : 6); // REX
1525 } else {
1526 // 32-bit
1527 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1528 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1529 int offset = ra_->reg2offset(dst_first);
1530 if (cbuf) {
1531 emit_opcode(*cbuf, 0xF3);
1532 if (Matcher::_regEncode[src_first] >= 8) {
1533 emit_opcode(*cbuf, Assembler::REX_R);
1534 }
1535 emit_opcode(*cbuf, 0x0F);
1536 emit_opcode(*cbuf, 0x11);
1537 encode_RegMem(*cbuf,
1538 Matcher::_regEncode[src_first],
1539 RSP_enc, 0x4, 0, offset,
1540 false);
1541 #ifndef PRODUCT
1542 } else if (!do_size) {
1543 st->print("movss [rsp + #%d], %s\t# spill",
1544 offset,
1545 Matcher::regName[src_first]);
1546 #endif
1547 }
1548 return
1549 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1550 ((Matcher::_regEncode[src_first] < 8)
1551 ? 5
1552 : 6); // REX
1553 }
1554 } else if (dst_first_rc == rc_int) {
1555 // xmm -> gpr
1556 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1557 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1558 // 64-bit
1559 if (cbuf) {
1560 emit_opcode(*cbuf, 0x66);
1561 if (Matcher::_regEncode[dst_first] < 8) {
1562 if (Matcher::_regEncode[src_first] < 8) {
1563 emit_opcode(*cbuf, Assembler::REX_W);
1564 } else {
1565 emit_opcode(*cbuf, Assembler::REX_WR); // attention!
1566 }
1567 } else {
1568 if (Matcher::_regEncode[src_first] < 8) {
1569 emit_opcode(*cbuf, Assembler::REX_WB); // attention!
1570 } else {
1571 emit_opcode(*cbuf, Assembler::REX_WRB);
1572 }
1573 }
1574 emit_opcode(*cbuf, 0x0F);
1575 emit_opcode(*cbuf, 0x7E);
1576 emit_rm(*cbuf, 0x3,
1577 Matcher::_regEncode[src_first] & 7,
1578 Matcher::_regEncode[dst_first] & 7);
1579 #ifndef PRODUCT
1580 } else if (!do_size) {
1581 st->print("movdq %s, %s\t# spill",
1582 Matcher::regName[dst_first],
1583 Matcher::regName[src_first]);
1584 #endif
1585 }
1586 return 5; // REX
1587 } else {
1588 // 32-bit
1589 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1590 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1591 if (cbuf) {
1592 emit_opcode(*cbuf, 0x66);
1593 if (Matcher::_regEncode[dst_first] < 8) {
1594 if (Matcher::_regEncode[src_first] >= 8) {
1595 emit_opcode(*cbuf, Assembler::REX_R); // attention!
1596 }
1597 } else {
1598 if (Matcher::_regEncode[src_first] < 8) {
1599 emit_opcode(*cbuf, Assembler::REX_B); // attention!
1600 } else {
1601 emit_opcode(*cbuf, Assembler::REX_RB);
1602 }
1603 }
1604 emit_opcode(*cbuf, 0x0F);
1605 emit_opcode(*cbuf, 0x7E);
1606 emit_rm(*cbuf, 0x3,
1607 Matcher::_regEncode[src_first] & 7,
1608 Matcher::_regEncode[dst_first] & 7);
1609 #ifndef PRODUCT
1610 } else if (!do_size) {
1611 st->print("movdl %s, %s\t# spill",
1612 Matcher::regName[dst_first],
1613 Matcher::regName[src_first]);
1614 #endif
1615 }
1616 return
1617 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1618 ? 4
1619 : 5; // REX
1620 }
1621 } else if (dst_first_rc == rc_float) {
1622 // xmm -> xmm
1623 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1624 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1625 // 64-bit
1626 if (cbuf) {
1627 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
1628 if (Matcher::_regEncode[dst_first] < 8) {
1629 if (Matcher::_regEncode[src_first] >= 8) {
1630 emit_opcode(*cbuf, Assembler::REX_B);
1631 }
1632 } else {
1633 if (Matcher::_regEncode[src_first] < 8) {
1634 emit_opcode(*cbuf, Assembler::REX_R);
1635 } else {
1636 emit_opcode(*cbuf, Assembler::REX_RB);
1637 }
1638 }
1639 emit_opcode(*cbuf, 0x0F);
1640 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1641 emit_rm(*cbuf, 0x3,
1642 Matcher::_regEncode[dst_first] & 7,
1643 Matcher::_regEncode[src_first] & 7);
1644 #ifndef PRODUCT
1645 } else if (!do_size) {
1646 st->print("%s %s, %s\t# spill",
1647 UseXmmRegToRegMoveAll ? "movapd" : "movsd ",
1648 Matcher::regName[dst_first],
1649 Matcher::regName[src_first]);
1650 #endif
1651 }
1652 return
1653 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1654 ? 4
1655 : 5; // REX
1656 } else {
1657 // 32-bit
1658 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1659 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1660 if (cbuf) {
1661 if (!UseXmmRegToRegMoveAll)
1662 emit_opcode(*cbuf, 0xF3);
1663 if (Matcher::_regEncode[dst_first] < 8) {
1664 if (Matcher::_regEncode[src_first] >= 8) {
1665 emit_opcode(*cbuf, Assembler::REX_B);
1666 }
1667 } else {
1668 if (Matcher::_regEncode[src_first] < 8) {
1669 emit_opcode(*cbuf, Assembler::REX_R);
1670 } else {
1671 emit_opcode(*cbuf, Assembler::REX_RB);
1672 }
1673 }
1674 emit_opcode(*cbuf, 0x0F);
1675 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1676 emit_rm(*cbuf, 0x3,
1677 Matcher::_regEncode[dst_first] & 7,
1678 Matcher::_regEncode[src_first] & 7);
1679 #ifndef PRODUCT
1680 } else if (!do_size) {
1681 st->print("%s %s, %s\t# spill",
1682 UseXmmRegToRegMoveAll ? "movaps" : "movss ",
1683 Matcher::regName[dst_first],
1684 Matcher::regName[src_first]);
1685 #endif
1686 }
1687 return
1688 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1689 ? (UseXmmRegToRegMoveAll ? 3 : 4)
1690 : (UseXmmRegToRegMoveAll ? 4 : 5); // REX
1691 }
1692 }
1693 }
1695 assert(0," foo ");
1696 Unimplemented();
1698 return 0;
1699 }
1701 #ifndef PRODUCT
1702 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const
1703 {
1704 implementation(NULL, ra_, false, st);
1705 }
1706 #endif
1708 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
1709 {
1710 implementation(&cbuf, ra_, false, NULL);
1711 }
1713 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const
1714 {
1715 return implementation(NULL, ra_, true, NULL);
1716 }
1718 //=============================================================================
1719 #ifndef PRODUCT
1720 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const
1721 {
1722 st->print("nop \t# %d bytes pad for loops and calls", _count);
1723 }
1724 #endif
1726 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const
1727 {
1728 MacroAssembler _masm(&cbuf);
1729 __ nop(_count);
1730 }
1732 uint MachNopNode::size(PhaseRegAlloc*) const
1733 {
1734 return _count;
1735 }
1738 //=============================================================================
1739 #ifndef PRODUCT
1740 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1741 {
1742 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1743 int reg = ra_->get_reg_first(this);
1744 st->print("leaq %s, [rsp + #%d]\t# box lock",
1745 Matcher::regName[reg], offset);
1746 }
1747 #endif
1749 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1750 {
1751 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1752 int reg = ra_->get_encode(this);
1753 if (offset >= 0x80) {
1754 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1755 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1756 emit_rm(cbuf, 0x2, reg & 7, 0x04);
1757 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1758 emit_d32(cbuf, offset);
1759 } else {
1760 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1761 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1762 emit_rm(cbuf, 0x1, reg & 7, 0x04);
1763 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1764 emit_d8(cbuf, offset);
1765 }
1766 }
1768 uint BoxLockNode::size(PhaseRegAlloc *ra_) const
1769 {
1770 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1771 return (offset < 0x80) ? 5 : 8; // REX
1772 }
1774 //=============================================================================
1776 // emit call stub, compiled java to interpreter
1777 void emit_java_to_interp(CodeBuffer& cbuf)
1778 {
1779 // Stub is fixed up when the corresponding call is converted from
1780 // calling compiled code to calling interpreted code.
1781 // movq rbx, 0
1782 // jmp -5 # to self
1784 address mark = cbuf.insts_mark(); // get mark within main instrs section
1786 // Note that the code buffer's insts_mark is always relative to insts.
1787 // That's why we must use the macroassembler to generate a stub.
1788 MacroAssembler _masm(&cbuf);
1790 address base =
1791 __ start_a_stub(Compile::MAX_stubs_size);
1792 if (base == NULL) return; // CodeBuffer::expand failed
1793 // static stub relocation stores the instruction address of the call
1794 __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64);
1795 // static stub relocation also tags the methodOop in the code-stream.
1796 __ movoop(rbx, (jobject) NULL); // method is zapped till fixup time
1797 // This is recognized as unresolved by relocs/nativeinst/ic code
1798 __ jump(RuntimeAddress(__ pc()));
1800 // Update current stubs pointer and restore insts_end.
1801 __ end_a_stub();
1802 }
1804 // size of call stub, compiled java to interpretor
1805 uint size_java_to_interp()
1806 {
1807 return 15; // movq (1+1+8); jmp (1+4)
1808 }
1810 // relocation entries for call stub, compiled java to interpretor
1811 uint reloc_java_to_interp()
1812 {
1813 return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
1814 }
1816 //=============================================================================
1817 #ifndef PRODUCT
1818 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1819 {
1820 if (UseCompressedOops) {
1821 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes()]\t# compressed klass");
1822 if (Universe::narrow_oop_shift() != 0) {
1823 st->print_cr("\tdecode_heap_oop_not_null rscratch1, rscratch1");
1824 }
1825 st->print_cr("\tcmpq rax, rscratch1\t # Inline cache check");
1826 } else {
1827 st->print_cr("\tcmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes()]\t"
1828 "# Inline cache check");
1829 }
1830 st->print_cr("\tjne SharedRuntime::_ic_miss_stub");
1831 st->print_cr("\tnop\t# nops to align entry point");
1832 }
1833 #endif
1835 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1836 {
1837 MacroAssembler masm(&cbuf);
1838 uint insts_size = cbuf.insts_size();
1839 if (UseCompressedOops) {
1840 masm.load_klass(rscratch1, j_rarg0);
1841 masm.cmpptr(rax, rscratch1);
1842 } else {
1843 masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
1844 }
1846 masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1848 /* WARNING these NOPs are critical so that verified entry point is properly
1849 4 bytes aligned for patching by NativeJump::patch_verified_entry() */
1850 int nops_cnt = 4 - ((cbuf.insts_size() - insts_size) & 0x3);
1851 if (OptoBreakpoint) {
1852 // Leave space for int3
1853 nops_cnt -= 1;
1854 }
1855 nops_cnt &= 0x3; // Do not add nops if code is aligned.
1856 if (nops_cnt > 0)
1857 masm.nop(nops_cnt);
1858 }
1860 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
1861 {
1862 return MachNode::size(ra_); // too many variables; just compute it
1863 // the hard way
1864 }
1867 //=============================================================================
1868 uint size_exception_handler()
1869 {
1870 // NativeCall instruction size is the same as NativeJump.
1871 // Note that this value is also credited (in output.cpp) to
1872 // the size of the code section.
1873 return NativeJump::instruction_size;
1874 }
1876 // Emit exception handler code.
1877 int emit_exception_handler(CodeBuffer& cbuf)
1878 {
1880 // Note that the code buffer's insts_mark is always relative to insts.
1881 // That's why we must use the macroassembler to generate a handler.
1882 MacroAssembler _masm(&cbuf);
1883 address base =
1884 __ start_a_stub(size_exception_handler());
1885 if (base == NULL) return 0; // CodeBuffer::expand failed
1886 int offset = __ offset();
1887 __ jump(RuntimeAddress(OptoRuntime::exception_blob()->entry_point()));
1888 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1889 __ end_a_stub();
1890 return offset;
1891 }
1893 uint size_deopt_handler()
1894 {
1895 // three 5 byte instructions
1896 return 15;
1897 }
1899 // Emit deopt handler code.
1900 int emit_deopt_handler(CodeBuffer& cbuf)
1901 {
1903 // Note that the code buffer's insts_mark is always relative to insts.
1904 // That's why we must use the macroassembler to generate a handler.
1905 MacroAssembler _masm(&cbuf);
1906 address base =
1907 __ start_a_stub(size_deopt_handler());
1908 if (base == NULL) return 0; // CodeBuffer::expand failed
1909 int offset = __ offset();
1910 address the_pc = (address) __ pc();
1911 Label next;
1912 // push a "the_pc" on the stack without destroying any registers
1913 // as they all may be live.
1915 // push address of "next"
1916 __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
1917 __ bind(next);
1918 // adjust it so it matches "the_pc"
1919 __ subptr(Address(rsp, 0), __ offset() - offset);
1920 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1921 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1922 __ end_a_stub();
1923 return offset;
1924 }
1927 const bool Matcher::match_rule_supported(int opcode) {
1928 if (!has_match_rule(opcode))
1929 return false;
1931 return true; // Per default match rules are supported.
1932 }
1934 int Matcher::regnum_to_fpu_offset(int regnum)
1935 {
1936 return regnum - 32; // The FP registers are in the second chunk
1937 }
1939 // This is UltraSparc specific, true just means we have fast l2f conversion
1940 const bool Matcher::convL2FSupported(void) {
1941 return true;
1942 }
1944 // Vector width in bytes
1945 const uint Matcher::vector_width_in_bytes(void) {
1946 return 8;
1947 }
1949 // Vector ideal reg
1950 const uint Matcher::vector_ideal_reg(void) {
1951 return Op_RegD;
1952 }
1954 // Is this branch offset short enough that a short branch can be used?
1955 //
1956 // NOTE: If the platform does not provide any short branch variants, then
1957 // this method should return false for offset 0.
1958 bool Matcher::is_short_branch_offset(int rule, int offset) {
1959 // the short version of jmpConUCF2 contains multiple branches,
1960 // making the reach slightly less
1961 if (rule == jmpConUCF2_rule)
1962 return (-126 <= offset && offset <= 125);
1963 return (-128 <= offset && offset <= 127);
1964 }
1966 const bool Matcher::isSimpleConstant64(jlong value) {
1967 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
1968 //return value == (int) value; // Cf. storeImmL and immL32.
1970 // Probably always true, even if a temp register is required.
1971 return true;
1972 }
1974 // The ecx parameter to rep stosq for the ClearArray node is in words.
1975 const bool Matcher::init_array_count_is_in_bytes = false;
1977 // Threshold size for cleararray.
1978 const int Matcher::init_array_short_size = 8 * BytesPerLong;
1980 // Should the Matcher clone shifts on addressing modes, expecting them
1981 // to be subsumed into complex addressing expressions or compute them
1982 // into registers? True for Intel but false for most RISCs
1983 const bool Matcher::clone_shift_expressions = true;
1985 // Do we need to mask the count passed to shift instructions or does
1986 // the cpu only look at the lower 5/6 bits anyway?
1987 const bool Matcher::need_masked_shift_count = false;
1989 bool Matcher::narrow_oop_use_complex_address() {
1990 assert(UseCompressedOops, "only for compressed oops code");
1991 return (LogMinObjAlignmentInBytes <= 3);
1992 }
1994 // Is it better to copy float constants, or load them directly from
1995 // memory? Intel can load a float constant from a direct address,
1996 // requiring no extra registers. Most RISCs will have to materialize
1997 // an address into a register first, so they would do better to copy
1998 // the constant from stack.
1999 const bool Matcher::rematerialize_float_constants = true; // XXX
2001 // If CPU can load and store mis-aligned doubles directly then no
2002 // fixup is needed. Else we split the double into 2 integer pieces
2003 // and move it piece-by-piece. Only happens when passing doubles into
2004 // C code as the Java calling convention forces doubles to be aligned.
2005 const bool Matcher::misaligned_doubles_ok = true;
2007 // No-op on amd64
2008 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
2010 // Advertise here if the CPU requires explicit rounding operations to
2011 // implement the UseStrictFP mode.
2012 const bool Matcher::strict_fp_requires_explicit_rounding = true;
2014 // Are floats conerted to double when stored to stack during deoptimization?
2015 // On x64 it is stored without convertion so we can use normal access.
2016 bool Matcher::float_in_double() { return false; }
2018 // Do ints take an entire long register or just half?
2019 const bool Matcher::int_in_long = true;
2021 // Return whether or not this register is ever used as an argument.
2022 // This function is used on startup to build the trampoline stubs in
2023 // generateOptoStub. Registers not mentioned will be killed by the VM
2024 // call in the trampoline, and arguments in those registers not be
2025 // available to the callee.
2026 bool Matcher::can_be_java_arg(int reg)
2027 {
2028 return
2029 reg == RDI_num || reg == RDI_H_num ||
2030 reg == RSI_num || reg == RSI_H_num ||
2031 reg == RDX_num || reg == RDX_H_num ||
2032 reg == RCX_num || reg == RCX_H_num ||
2033 reg == R8_num || reg == R8_H_num ||
2034 reg == R9_num || reg == R9_H_num ||
2035 reg == R12_num || reg == R12_H_num ||
2036 reg == XMM0_num || reg == XMM0_H_num ||
2037 reg == XMM1_num || reg == XMM1_H_num ||
2038 reg == XMM2_num || reg == XMM2_H_num ||
2039 reg == XMM3_num || reg == XMM3_H_num ||
2040 reg == XMM4_num || reg == XMM4_H_num ||
2041 reg == XMM5_num || reg == XMM5_H_num ||
2042 reg == XMM6_num || reg == XMM6_H_num ||
2043 reg == XMM7_num || reg == XMM7_H_num;
2044 }
2046 bool Matcher::is_spillable_arg(int reg)
2047 {
2048 return can_be_java_arg(reg);
2049 }
2051 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
2052 // In 64 bit mode a code which use multiply when
2053 // devisor is constant is faster than hardware
2054 // DIV instruction (it uses MulHiL).
2055 return false;
2056 }
2058 // Register for DIVI projection of divmodI
2059 RegMask Matcher::divI_proj_mask() {
2060 return INT_RAX_REG_mask;
2061 }
2063 // Register for MODI projection of divmodI
2064 RegMask Matcher::modI_proj_mask() {
2065 return INT_RDX_REG_mask;
2066 }
2068 // Register for DIVL projection of divmodL
2069 RegMask Matcher::divL_proj_mask() {
2070 return LONG_RAX_REG_mask;
2071 }
2073 // Register for MODL projection of divmodL
2074 RegMask Matcher::modL_proj_mask() {
2075 return LONG_RDX_REG_mask;
2076 }
2078 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
2079 return PTR_RBP_REG_mask;
2080 }
2082 static Address build_address(int b, int i, int s, int d) {
2083 Register index = as_Register(i);
2084 Address::ScaleFactor scale = (Address::ScaleFactor)s;
2085 if (index == rsp) {
2086 index = noreg;
2087 scale = Address::no_scale;
2088 }
2089 Address addr(as_Register(b), index, scale, d);
2090 return addr;
2091 }
2093 %}
2095 //----------ENCODING BLOCK-----------------------------------------------------
2096 // This block specifies the encoding classes used by the compiler to
2097 // output byte streams. Encoding classes are parameterized macros
2098 // used by Machine Instruction Nodes in order to generate the bit
2099 // encoding of the instruction. Operands specify their base encoding
2100 // interface with the interface keyword. There are currently
2101 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
2102 // COND_INTER. REG_INTER causes an operand to generate a function
2103 // which returns its register number when queried. CONST_INTER causes
2104 // an operand to generate a function which returns the value of the
2105 // constant when queried. MEMORY_INTER causes an operand to generate
2106 // four functions which return the Base Register, the Index Register,
2107 // the Scale Value, and the Offset Value of the operand when queried.
2108 // COND_INTER causes an operand to generate six functions which return
2109 // the encoding code (ie - encoding bits for the instruction)
2110 // associated with each basic boolean condition for a conditional
2111 // instruction.
2112 //
2113 // Instructions specify two basic values for encoding. Again, a
2114 // function is available to check if the constant displacement is an
2115 // oop. They use the ins_encode keyword to specify their encoding
2116 // classes (which must be a sequence of enc_class names, and their
2117 // parameters, specified in the encoding block), and they use the
2118 // opcode keyword to specify, in order, their primary, secondary, and
2119 // tertiary opcode. Only the opcode sections which a particular
2120 // instruction needs for encoding need to be specified.
2121 encode %{
2122 // Build emit functions for each basic byte or larger field in the
2123 // intel encoding scheme (opcode, rm, sib, immediate), and call them
2124 // from C++ code in the enc_class source block. Emit functions will
2125 // live in the main source block for now. In future, we can
2126 // generalize this by adding a syntax that specifies the sizes of
2127 // fields in an order, so that the adlc can build the emit functions
2128 // automagically
2130 // Emit primary opcode
2131 enc_class OpcP
2132 %{
2133 emit_opcode(cbuf, $primary);
2134 %}
2136 // Emit secondary opcode
2137 enc_class OpcS
2138 %{
2139 emit_opcode(cbuf, $secondary);
2140 %}
2142 // Emit tertiary opcode
2143 enc_class OpcT
2144 %{
2145 emit_opcode(cbuf, $tertiary);
2146 %}
2148 // Emit opcode directly
2149 enc_class Opcode(immI d8)
2150 %{
2151 emit_opcode(cbuf, $d8$$constant);
2152 %}
2154 // Emit size prefix
2155 enc_class SizePrefix
2156 %{
2157 emit_opcode(cbuf, 0x66);
2158 %}
2160 enc_class reg(rRegI reg)
2161 %{
2162 emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
2163 %}
2165 enc_class reg_reg(rRegI dst, rRegI src)
2166 %{
2167 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2168 %}
2170 enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
2171 %{
2172 emit_opcode(cbuf, $opcode$$constant);
2173 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2174 %}
2176 enc_class cmpfp_fixup()
2177 %{
2178 // jnp,s exit
2179 emit_opcode(cbuf, 0x7B);
2180 emit_d8(cbuf, 0x0A);
2182 // pushfq
2183 emit_opcode(cbuf, 0x9C);
2185 // andq $0xffffff2b, (%rsp)
2186 emit_opcode(cbuf, Assembler::REX_W);
2187 emit_opcode(cbuf, 0x81);
2188 emit_opcode(cbuf, 0x24);
2189 emit_opcode(cbuf, 0x24);
2190 emit_d32(cbuf, 0xffffff2b);
2192 // popfq
2193 emit_opcode(cbuf, 0x9D);
2195 // nop (target for branch to avoid branch to branch)
2196 emit_opcode(cbuf, 0x90);
2197 %}
2199 enc_class cmpfp3(rRegI dst)
2200 %{
2201 int dstenc = $dst$$reg;
2203 // movl $dst, -1
2204 if (dstenc >= 8) {
2205 emit_opcode(cbuf, Assembler::REX_B);
2206 }
2207 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
2208 emit_d32(cbuf, -1);
2210 // jp,s done
2211 emit_opcode(cbuf, 0x7A);
2212 emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
2214 // jb,s done
2215 emit_opcode(cbuf, 0x72);
2216 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
2218 // setne $dst
2219 if (dstenc >= 4) {
2220 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
2221 }
2222 emit_opcode(cbuf, 0x0F);
2223 emit_opcode(cbuf, 0x95);
2224 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
2226 // movzbl $dst, $dst
2227 if (dstenc >= 4) {
2228 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
2229 }
2230 emit_opcode(cbuf, 0x0F);
2231 emit_opcode(cbuf, 0xB6);
2232 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
2233 %}
2235 enc_class cdql_enc(no_rax_rdx_RegI div)
2236 %{
2237 // Full implementation of Java idiv and irem; checks for
2238 // special case as described in JVM spec., p.243 & p.271.
2239 //
2240 // normal case special case
2241 //
2242 // input : rax: dividend min_int
2243 // reg: divisor -1
2244 //
2245 // output: rax: quotient (= rax idiv reg) min_int
2246 // rdx: remainder (= rax irem reg) 0
2247 //
2248 // Code sequnce:
2249 //
2250 // 0: 3d 00 00 00 80 cmp $0x80000000,%eax
2251 // 5: 75 07/08 jne e <normal>
2252 // 7: 33 d2 xor %edx,%edx
2253 // [div >= 8 -> offset + 1]
2254 // [REX_B]
2255 // 9: 83 f9 ff cmp $0xffffffffffffffff,$div
2256 // c: 74 03/04 je 11 <done>
2257 // 000000000000000e <normal>:
2258 // e: 99 cltd
2259 // [div >= 8 -> offset + 1]
2260 // [REX_B]
2261 // f: f7 f9 idiv $div
2262 // 0000000000000011 <done>:
2264 // cmp $0x80000000,%eax
2265 emit_opcode(cbuf, 0x3d);
2266 emit_d8(cbuf, 0x00);
2267 emit_d8(cbuf, 0x00);
2268 emit_d8(cbuf, 0x00);
2269 emit_d8(cbuf, 0x80);
2271 // jne e <normal>
2272 emit_opcode(cbuf, 0x75);
2273 emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
2275 // xor %edx,%edx
2276 emit_opcode(cbuf, 0x33);
2277 emit_d8(cbuf, 0xD2);
2279 // cmp $0xffffffffffffffff,%ecx
2280 if ($div$$reg >= 8) {
2281 emit_opcode(cbuf, Assembler::REX_B);
2282 }
2283 emit_opcode(cbuf, 0x83);
2284 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2285 emit_d8(cbuf, 0xFF);
2287 // je 11 <done>
2288 emit_opcode(cbuf, 0x74);
2289 emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
2291 // <normal>
2292 // cltd
2293 emit_opcode(cbuf, 0x99);
2295 // idivl (note: must be emitted by the user of this rule)
2296 // <done>
2297 %}
2299 enc_class cdqq_enc(no_rax_rdx_RegL div)
2300 %{
2301 // Full implementation of Java ldiv and lrem; checks for
2302 // special case as described in JVM spec., p.243 & p.271.
2303 //
2304 // normal case special case
2305 //
2306 // input : rax: dividend min_long
2307 // reg: divisor -1
2308 //
2309 // output: rax: quotient (= rax idiv reg) min_long
2310 // rdx: remainder (= rax irem reg) 0
2311 //
2312 // Code sequnce:
2313 //
2314 // 0: 48 ba 00 00 00 00 00 mov $0x8000000000000000,%rdx
2315 // 7: 00 00 80
2316 // a: 48 39 d0 cmp %rdx,%rax
2317 // d: 75 08 jne 17 <normal>
2318 // f: 33 d2 xor %edx,%edx
2319 // 11: 48 83 f9 ff cmp $0xffffffffffffffff,$div
2320 // 15: 74 05 je 1c <done>
2321 // 0000000000000017 <normal>:
2322 // 17: 48 99 cqto
2323 // 19: 48 f7 f9 idiv $div
2324 // 000000000000001c <done>:
2326 // mov $0x8000000000000000,%rdx
2327 emit_opcode(cbuf, Assembler::REX_W);
2328 emit_opcode(cbuf, 0xBA);
2329 emit_d8(cbuf, 0x00);
2330 emit_d8(cbuf, 0x00);
2331 emit_d8(cbuf, 0x00);
2332 emit_d8(cbuf, 0x00);
2333 emit_d8(cbuf, 0x00);
2334 emit_d8(cbuf, 0x00);
2335 emit_d8(cbuf, 0x00);
2336 emit_d8(cbuf, 0x80);
2338 // cmp %rdx,%rax
2339 emit_opcode(cbuf, Assembler::REX_W);
2340 emit_opcode(cbuf, 0x39);
2341 emit_d8(cbuf, 0xD0);
2343 // jne 17 <normal>
2344 emit_opcode(cbuf, 0x75);
2345 emit_d8(cbuf, 0x08);
2347 // xor %edx,%edx
2348 emit_opcode(cbuf, 0x33);
2349 emit_d8(cbuf, 0xD2);
2351 // cmp $0xffffffffffffffff,$div
2352 emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
2353 emit_opcode(cbuf, 0x83);
2354 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2355 emit_d8(cbuf, 0xFF);
2357 // je 1e <done>
2358 emit_opcode(cbuf, 0x74);
2359 emit_d8(cbuf, 0x05);
2361 // <normal>
2362 // cqto
2363 emit_opcode(cbuf, Assembler::REX_W);
2364 emit_opcode(cbuf, 0x99);
2366 // idivq (note: must be emitted by the user of this rule)
2367 // <done>
2368 %}
2370 // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
2371 enc_class OpcSE(immI imm)
2372 %{
2373 // Emit primary opcode and set sign-extend bit
2374 // Check for 8-bit immediate, and set sign extend bit in opcode
2375 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2376 emit_opcode(cbuf, $primary | 0x02);
2377 } else {
2378 // 32-bit immediate
2379 emit_opcode(cbuf, $primary);
2380 }
2381 %}
2383 enc_class OpcSErm(rRegI dst, immI imm)
2384 %{
2385 // OpcSEr/m
2386 int dstenc = $dst$$reg;
2387 if (dstenc >= 8) {
2388 emit_opcode(cbuf, Assembler::REX_B);
2389 dstenc -= 8;
2390 }
2391 // Emit primary opcode and set sign-extend bit
2392 // Check for 8-bit immediate, and set sign extend bit in opcode
2393 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2394 emit_opcode(cbuf, $primary | 0x02);
2395 } else {
2396 // 32-bit immediate
2397 emit_opcode(cbuf, $primary);
2398 }
2399 // Emit r/m byte with secondary opcode, after primary opcode.
2400 emit_rm(cbuf, 0x3, $secondary, dstenc);
2401 %}
2403 enc_class OpcSErm_wide(rRegL dst, immI imm)
2404 %{
2405 // OpcSEr/m
2406 int dstenc = $dst$$reg;
2407 if (dstenc < 8) {
2408 emit_opcode(cbuf, Assembler::REX_W);
2409 } else {
2410 emit_opcode(cbuf, Assembler::REX_WB);
2411 dstenc -= 8;
2412 }
2413 // Emit primary opcode and set sign-extend bit
2414 // Check for 8-bit immediate, and set sign extend bit in opcode
2415 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2416 emit_opcode(cbuf, $primary | 0x02);
2417 } else {
2418 // 32-bit immediate
2419 emit_opcode(cbuf, $primary);
2420 }
2421 // Emit r/m byte with secondary opcode, after primary opcode.
2422 emit_rm(cbuf, 0x3, $secondary, dstenc);
2423 %}
2425 enc_class Con8or32(immI imm)
2426 %{
2427 // Check for 8-bit immediate, and set sign extend bit in opcode
2428 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2429 $$$emit8$imm$$constant;
2430 } else {
2431 // 32-bit immediate
2432 $$$emit32$imm$$constant;
2433 }
2434 %}
2436 enc_class Lbl(label labl)
2437 %{
2438 // JMP, CALL
2439 Label* l = $labl$$label;
2440 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.insts_size() + 4)) : 0);
2441 %}
2443 enc_class LblShort(label labl)
2444 %{
2445 // JMP, CALL
2446 Label* l = $labl$$label;
2447 int disp = l ? (l->loc_pos() - (cbuf.insts_size() + 1)) : 0;
2448 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2449 emit_d8(cbuf, disp);
2450 %}
2452 enc_class opc2_reg(rRegI dst)
2453 %{
2454 // BSWAP
2455 emit_cc(cbuf, $secondary, $dst$$reg);
2456 %}
2458 enc_class opc3_reg(rRegI dst)
2459 %{
2460 // BSWAP
2461 emit_cc(cbuf, $tertiary, $dst$$reg);
2462 %}
2464 enc_class reg_opc(rRegI div)
2465 %{
2466 // INC, DEC, IDIV, IMOD, JMP indirect, ...
2467 emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
2468 %}
2470 enc_class Jcc(cmpOp cop, label labl)
2471 %{
2472 // JCC
2473 Label* l = $labl$$label;
2474 $$$emit8$primary;
2475 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2476 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.insts_size() + 4)) : 0);
2477 %}
2479 enc_class JccShort (cmpOp cop, label labl)
2480 %{
2481 // JCC
2482 Label *l = $labl$$label;
2483 emit_cc(cbuf, $primary, $cop$$cmpcode);
2484 int disp = l ? (l->loc_pos() - (cbuf.insts_size() + 1)) : 0;
2485 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2486 emit_d8(cbuf, disp);
2487 %}
2489 enc_class enc_cmov(cmpOp cop)
2490 %{
2491 // CMOV
2492 $$$emit8$primary;
2493 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2494 %}
2496 enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
2497 %{
2498 // Invert sense of branch from sense of cmov
2499 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2500 emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
2501 ? (UseXmmRegToRegMoveAll ? 3 : 4)
2502 : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
2503 // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
2504 if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
2505 if ($dst$$reg < 8) {
2506 if ($src$$reg >= 8) {
2507 emit_opcode(cbuf, Assembler::REX_B);
2508 }
2509 } else {
2510 if ($src$$reg < 8) {
2511 emit_opcode(cbuf, Assembler::REX_R);
2512 } else {
2513 emit_opcode(cbuf, Assembler::REX_RB);
2514 }
2515 }
2516 emit_opcode(cbuf, 0x0F);
2517 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2518 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2519 %}
2521 enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
2522 %{
2523 // Invert sense of branch from sense of cmov
2524 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2525 emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
2527 // UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
2528 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
2529 if ($dst$$reg < 8) {
2530 if ($src$$reg >= 8) {
2531 emit_opcode(cbuf, Assembler::REX_B);
2532 }
2533 } else {
2534 if ($src$$reg < 8) {
2535 emit_opcode(cbuf, Assembler::REX_R);
2536 } else {
2537 emit_opcode(cbuf, Assembler::REX_RB);
2538 }
2539 }
2540 emit_opcode(cbuf, 0x0F);
2541 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2542 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2543 %}
2545 enc_class enc_PartialSubtypeCheck()
2546 %{
2547 Register Rrdi = as_Register(RDI_enc); // result register
2548 Register Rrax = as_Register(RAX_enc); // super class
2549 Register Rrcx = as_Register(RCX_enc); // killed
2550 Register Rrsi = as_Register(RSI_enc); // sub class
2551 Label miss;
2552 const bool set_cond_codes = true;
2554 MacroAssembler _masm(&cbuf);
2555 __ check_klass_subtype_slow_path(Rrsi, Rrax, Rrcx, Rrdi,
2556 NULL, &miss,
2557 /*set_cond_codes:*/ true);
2558 if ($primary) {
2559 __ xorptr(Rrdi, Rrdi);
2560 }
2561 __ bind(miss);
2562 %}
2564 enc_class Java_To_Interpreter(method meth)
2565 %{
2566 // CALL Java_To_Interpreter
2567 // This is the instruction starting address for relocation info.
2568 cbuf.set_insts_mark();
2569 $$$emit8$primary;
2570 // CALL directly to the runtime
2571 emit_d32_reloc(cbuf,
2572 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2573 runtime_call_Relocation::spec(),
2574 RELOC_DISP32);
2575 %}
2577 enc_class preserve_SP %{
2578 debug_only(int off0 = cbuf.insts_size());
2579 MacroAssembler _masm(&cbuf);
2580 // RBP is preserved across all calls, even compiled calls.
2581 // Use it to preserve RSP in places where the callee might change the SP.
2582 __ movptr(rbp_mh_SP_save, rsp);
2583 debug_only(int off1 = cbuf.insts_size());
2584 assert(off1 - off0 == preserve_SP_size(), "correct size prediction");
2585 %}
2587 enc_class restore_SP %{
2588 MacroAssembler _masm(&cbuf);
2589 __ movptr(rsp, rbp_mh_SP_save);
2590 %}
2592 enc_class Java_Static_Call(method meth)
2593 %{
2594 // JAVA STATIC CALL
2595 // CALL to fixup routine. Fixup routine uses ScopeDesc info to
2596 // determine who we intended to call.
2597 cbuf.set_insts_mark();
2598 $$$emit8$primary;
2600 if (!_method) {
2601 emit_d32_reloc(cbuf,
2602 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2603 runtime_call_Relocation::spec(),
2604 RELOC_DISP32);
2605 } else if (_optimized_virtual) {
2606 emit_d32_reloc(cbuf,
2607 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2608 opt_virtual_call_Relocation::spec(),
2609 RELOC_DISP32);
2610 } else {
2611 emit_d32_reloc(cbuf,
2612 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2613 static_call_Relocation::spec(),
2614 RELOC_DISP32);
2615 }
2616 if (_method) {
2617 // Emit stub for static call
2618 emit_java_to_interp(cbuf);
2619 }
2620 %}
2622 enc_class Java_Dynamic_Call(method meth)
2623 %{
2624 // JAVA DYNAMIC CALL
2625 // !!!!!
2626 // Generate "movq rax, -1", placeholder instruction to load oop-info
2627 // emit_call_dynamic_prologue( cbuf );
2628 cbuf.set_insts_mark();
2630 // movq rax, -1
2631 emit_opcode(cbuf, Assembler::REX_W);
2632 emit_opcode(cbuf, 0xB8 | RAX_enc);
2633 emit_d64_reloc(cbuf,
2634 (int64_t) Universe::non_oop_word(),
2635 oop_Relocation::spec_for_immediate(), RELOC_IMM64);
2636 address virtual_call_oop_addr = cbuf.insts_mark();
2637 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2638 // who we intended to call.
2639 cbuf.set_insts_mark();
2640 $$$emit8$primary;
2641 emit_d32_reloc(cbuf,
2642 (int) ($meth$$method - ((intptr_t) cbuf.insts_end()) - 4),
2643 virtual_call_Relocation::spec(virtual_call_oop_addr),
2644 RELOC_DISP32);
2645 %}
2647 enc_class Java_Compiled_Call(method meth)
2648 %{
2649 // JAVA COMPILED CALL
2650 int disp = in_bytes(methodOopDesc:: from_compiled_offset());
2652 // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
2653 // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
2655 // callq *disp(%rax)
2656 cbuf.set_insts_mark();
2657 $$$emit8$primary;
2658 if (disp < 0x80) {
2659 emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
2660 emit_d8(cbuf, disp); // Displacement
2661 } else {
2662 emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
2663 emit_d32(cbuf, disp); // Displacement
2664 }
2665 %}
2667 enc_class reg_opc_imm(rRegI dst, immI8 shift)
2668 %{
2669 // SAL, SAR, SHR
2670 int dstenc = $dst$$reg;
2671 if (dstenc >= 8) {
2672 emit_opcode(cbuf, Assembler::REX_B);
2673 dstenc -= 8;
2674 }
2675 $$$emit8$primary;
2676 emit_rm(cbuf, 0x3, $secondary, dstenc);
2677 $$$emit8$shift$$constant;
2678 %}
2680 enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
2681 %{
2682 // SAL, SAR, SHR
2683 int dstenc = $dst$$reg;
2684 if (dstenc < 8) {
2685 emit_opcode(cbuf, Assembler::REX_W);
2686 } else {
2687 emit_opcode(cbuf, Assembler::REX_WB);
2688 dstenc -= 8;
2689 }
2690 $$$emit8$primary;
2691 emit_rm(cbuf, 0x3, $secondary, dstenc);
2692 $$$emit8$shift$$constant;
2693 %}
2695 enc_class load_immI(rRegI dst, immI src)
2696 %{
2697 int dstenc = $dst$$reg;
2698 if (dstenc >= 8) {
2699 emit_opcode(cbuf, Assembler::REX_B);
2700 dstenc -= 8;
2701 }
2702 emit_opcode(cbuf, 0xB8 | dstenc);
2703 $$$emit32$src$$constant;
2704 %}
2706 enc_class load_immL(rRegL dst, immL src)
2707 %{
2708 int dstenc = $dst$$reg;
2709 if (dstenc < 8) {
2710 emit_opcode(cbuf, Assembler::REX_W);
2711 } else {
2712 emit_opcode(cbuf, Assembler::REX_WB);
2713 dstenc -= 8;
2714 }
2715 emit_opcode(cbuf, 0xB8 | dstenc);
2716 emit_d64(cbuf, $src$$constant);
2717 %}
2719 enc_class load_immUL32(rRegL dst, immUL32 src)
2720 %{
2721 // same as load_immI, but this time we care about zeroes in the high word
2722 int dstenc = $dst$$reg;
2723 if (dstenc >= 8) {
2724 emit_opcode(cbuf, Assembler::REX_B);
2725 dstenc -= 8;
2726 }
2727 emit_opcode(cbuf, 0xB8 | dstenc);
2728 $$$emit32$src$$constant;
2729 %}
2731 enc_class load_immL32(rRegL dst, immL32 src)
2732 %{
2733 int dstenc = $dst$$reg;
2734 if (dstenc < 8) {
2735 emit_opcode(cbuf, Assembler::REX_W);
2736 } else {
2737 emit_opcode(cbuf, Assembler::REX_WB);
2738 dstenc -= 8;
2739 }
2740 emit_opcode(cbuf, 0xC7);
2741 emit_rm(cbuf, 0x03, 0x00, dstenc);
2742 $$$emit32$src$$constant;
2743 %}
2745 enc_class load_immP31(rRegP dst, immP32 src)
2746 %{
2747 // same as load_immI, but this time we care about zeroes in the high word
2748 int dstenc = $dst$$reg;
2749 if (dstenc >= 8) {
2750 emit_opcode(cbuf, Assembler::REX_B);
2751 dstenc -= 8;
2752 }
2753 emit_opcode(cbuf, 0xB8 | dstenc);
2754 $$$emit32$src$$constant;
2755 %}
2757 enc_class load_immP(rRegP dst, immP src)
2758 %{
2759 int dstenc = $dst$$reg;
2760 if (dstenc < 8) {
2761 emit_opcode(cbuf, Assembler::REX_W);
2762 } else {
2763 emit_opcode(cbuf, Assembler::REX_WB);
2764 dstenc -= 8;
2765 }
2766 emit_opcode(cbuf, 0xB8 | dstenc);
2767 // This next line should be generated from ADLC
2768 if ($src->constant_is_oop()) {
2769 emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
2770 } else {
2771 emit_d64(cbuf, $src$$constant);
2772 }
2773 %}
2775 // Encode a reg-reg copy. If it is useless, then empty encoding.
2776 enc_class enc_copy(rRegI dst, rRegI src)
2777 %{
2778 encode_copy(cbuf, $dst$$reg, $src$$reg);
2779 %}
2781 // Encode xmm reg-reg copy. If it is useless, then empty encoding.
2782 enc_class enc_CopyXD( RegD dst, RegD src ) %{
2783 encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
2784 %}
2786 enc_class enc_copy_always(rRegI dst, rRegI src)
2787 %{
2788 int srcenc = $src$$reg;
2789 int dstenc = $dst$$reg;
2791 if (dstenc < 8) {
2792 if (srcenc >= 8) {
2793 emit_opcode(cbuf, Assembler::REX_B);
2794 srcenc -= 8;
2795 }
2796 } else {
2797 if (srcenc < 8) {
2798 emit_opcode(cbuf, Assembler::REX_R);
2799 } else {
2800 emit_opcode(cbuf, Assembler::REX_RB);
2801 srcenc -= 8;
2802 }
2803 dstenc -= 8;
2804 }
2806 emit_opcode(cbuf, 0x8B);
2807 emit_rm(cbuf, 0x3, dstenc, srcenc);
2808 %}
2810 enc_class enc_copy_wide(rRegL dst, rRegL src)
2811 %{
2812 int srcenc = $src$$reg;
2813 int dstenc = $dst$$reg;
2815 if (dstenc != srcenc) {
2816 if (dstenc < 8) {
2817 if (srcenc < 8) {
2818 emit_opcode(cbuf, Assembler::REX_W);
2819 } else {
2820 emit_opcode(cbuf, Assembler::REX_WB);
2821 srcenc -= 8;
2822 }
2823 } else {
2824 if (srcenc < 8) {
2825 emit_opcode(cbuf, Assembler::REX_WR);
2826 } else {
2827 emit_opcode(cbuf, Assembler::REX_WRB);
2828 srcenc -= 8;
2829 }
2830 dstenc -= 8;
2831 }
2832 emit_opcode(cbuf, 0x8B);
2833 emit_rm(cbuf, 0x3, dstenc, srcenc);
2834 }
2835 %}
2837 enc_class Con32(immI src)
2838 %{
2839 // Output immediate
2840 $$$emit32$src$$constant;
2841 %}
2843 enc_class Con64(immL src)
2844 %{
2845 // Output immediate
2846 emit_d64($src$$constant);
2847 %}
2849 enc_class Con32F_as_bits(immF src)
2850 %{
2851 // Output Float immediate bits
2852 jfloat jf = $src$$constant;
2853 jint jf_as_bits = jint_cast(jf);
2854 emit_d32(cbuf, jf_as_bits);
2855 %}
2857 enc_class Con16(immI src)
2858 %{
2859 // Output immediate
2860 $$$emit16$src$$constant;
2861 %}
2863 // How is this different from Con32??? XXX
2864 enc_class Con_d32(immI src)
2865 %{
2866 emit_d32(cbuf,$src$$constant);
2867 %}
2869 enc_class conmemref (rRegP t1) %{ // Con32(storeImmI)
2870 // Output immediate memory reference
2871 emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2872 emit_d32(cbuf, 0x00);
2873 %}
2875 enc_class lock_prefix()
2876 %{
2877 if (os::is_MP()) {
2878 emit_opcode(cbuf, 0xF0); // lock
2879 }
2880 %}
2882 enc_class REX_mem(memory mem)
2883 %{
2884 if ($mem$$base >= 8) {
2885 if ($mem$$index < 8) {
2886 emit_opcode(cbuf, Assembler::REX_B);
2887 } else {
2888 emit_opcode(cbuf, Assembler::REX_XB);
2889 }
2890 } else {
2891 if ($mem$$index >= 8) {
2892 emit_opcode(cbuf, Assembler::REX_X);
2893 }
2894 }
2895 %}
2897 enc_class REX_mem_wide(memory mem)
2898 %{
2899 if ($mem$$base >= 8) {
2900 if ($mem$$index < 8) {
2901 emit_opcode(cbuf, Assembler::REX_WB);
2902 } else {
2903 emit_opcode(cbuf, Assembler::REX_WXB);
2904 }
2905 } else {
2906 if ($mem$$index < 8) {
2907 emit_opcode(cbuf, Assembler::REX_W);
2908 } else {
2909 emit_opcode(cbuf, Assembler::REX_WX);
2910 }
2911 }
2912 %}
2914 // for byte regs
2915 enc_class REX_breg(rRegI reg)
2916 %{
2917 if ($reg$$reg >= 4) {
2918 emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
2919 }
2920 %}
2922 // for byte regs
2923 enc_class REX_reg_breg(rRegI dst, rRegI src)
2924 %{
2925 if ($dst$$reg < 8) {
2926 if ($src$$reg >= 4) {
2927 emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
2928 }
2929 } else {
2930 if ($src$$reg < 8) {
2931 emit_opcode(cbuf, Assembler::REX_R);
2932 } else {
2933 emit_opcode(cbuf, Assembler::REX_RB);
2934 }
2935 }
2936 %}
2938 // for byte regs
2939 enc_class REX_breg_mem(rRegI reg, memory mem)
2940 %{
2941 if ($reg$$reg < 8) {
2942 if ($mem$$base < 8) {
2943 if ($mem$$index >= 8) {
2944 emit_opcode(cbuf, Assembler::REX_X);
2945 } else if ($reg$$reg >= 4) {
2946 emit_opcode(cbuf, Assembler::REX);
2947 }
2948 } else {
2949 if ($mem$$index < 8) {
2950 emit_opcode(cbuf, Assembler::REX_B);
2951 } else {
2952 emit_opcode(cbuf, Assembler::REX_XB);
2953 }
2954 }
2955 } else {
2956 if ($mem$$base < 8) {
2957 if ($mem$$index < 8) {
2958 emit_opcode(cbuf, Assembler::REX_R);
2959 } else {
2960 emit_opcode(cbuf, Assembler::REX_RX);
2961 }
2962 } else {
2963 if ($mem$$index < 8) {
2964 emit_opcode(cbuf, Assembler::REX_RB);
2965 } else {
2966 emit_opcode(cbuf, Assembler::REX_RXB);
2967 }
2968 }
2969 }
2970 %}
2972 enc_class REX_reg(rRegI reg)
2973 %{
2974 if ($reg$$reg >= 8) {
2975 emit_opcode(cbuf, Assembler::REX_B);
2976 }
2977 %}
2979 enc_class REX_reg_wide(rRegI reg)
2980 %{
2981 if ($reg$$reg < 8) {
2982 emit_opcode(cbuf, Assembler::REX_W);
2983 } else {
2984 emit_opcode(cbuf, Assembler::REX_WB);
2985 }
2986 %}
2988 enc_class REX_reg_reg(rRegI dst, rRegI src)
2989 %{
2990 if ($dst$$reg < 8) {
2991 if ($src$$reg >= 8) {
2992 emit_opcode(cbuf, Assembler::REX_B);
2993 }
2994 } else {
2995 if ($src$$reg < 8) {
2996 emit_opcode(cbuf, Assembler::REX_R);
2997 } else {
2998 emit_opcode(cbuf, Assembler::REX_RB);
2999 }
3000 }
3001 %}
3003 enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
3004 %{
3005 if ($dst$$reg < 8) {
3006 if ($src$$reg < 8) {
3007 emit_opcode(cbuf, Assembler::REX_W);
3008 } else {
3009 emit_opcode(cbuf, Assembler::REX_WB);
3010 }
3011 } else {
3012 if ($src$$reg < 8) {
3013 emit_opcode(cbuf, Assembler::REX_WR);
3014 } else {
3015 emit_opcode(cbuf, Assembler::REX_WRB);
3016 }
3017 }
3018 %}
3020 enc_class REX_reg_mem(rRegI reg, memory mem)
3021 %{
3022 if ($reg$$reg < 8) {
3023 if ($mem$$base < 8) {
3024 if ($mem$$index >= 8) {
3025 emit_opcode(cbuf, Assembler::REX_X);
3026 }
3027 } else {
3028 if ($mem$$index < 8) {
3029 emit_opcode(cbuf, Assembler::REX_B);
3030 } else {
3031 emit_opcode(cbuf, Assembler::REX_XB);
3032 }
3033 }
3034 } else {
3035 if ($mem$$base < 8) {
3036 if ($mem$$index < 8) {
3037 emit_opcode(cbuf, Assembler::REX_R);
3038 } else {
3039 emit_opcode(cbuf, Assembler::REX_RX);
3040 }
3041 } else {
3042 if ($mem$$index < 8) {
3043 emit_opcode(cbuf, Assembler::REX_RB);
3044 } else {
3045 emit_opcode(cbuf, Assembler::REX_RXB);
3046 }
3047 }
3048 }
3049 %}
3051 enc_class REX_reg_mem_wide(rRegL reg, memory mem)
3052 %{
3053 if ($reg$$reg < 8) {
3054 if ($mem$$base < 8) {
3055 if ($mem$$index < 8) {
3056 emit_opcode(cbuf, Assembler::REX_W);
3057 } else {
3058 emit_opcode(cbuf, Assembler::REX_WX);
3059 }
3060 } else {
3061 if ($mem$$index < 8) {
3062 emit_opcode(cbuf, Assembler::REX_WB);
3063 } else {
3064 emit_opcode(cbuf, Assembler::REX_WXB);
3065 }
3066 }
3067 } else {
3068 if ($mem$$base < 8) {
3069 if ($mem$$index < 8) {
3070 emit_opcode(cbuf, Assembler::REX_WR);
3071 } else {
3072 emit_opcode(cbuf, Assembler::REX_WRX);
3073 }
3074 } else {
3075 if ($mem$$index < 8) {
3076 emit_opcode(cbuf, Assembler::REX_WRB);
3077 } else {
3078 emit_opcode(cbuf, Assembler::REX_WRXB);
3079 }
3080 }
3081 }
3082 %}
3084 enc_class reg_mem(rRegI ereg, memory mem)
3085 %{
3086 // High registers handle in encode_RegMem
3087 int reg = $ereg$$reg;
3088 int base = $mem$$base;
3089 int index = $mem$$index;
3090 int scale = $mem$$scale;
3091 int disp = $mem$$disp;
3092 bool disp_is_oop = $mem->disp_is_oop();
3094 encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
3095 %}
3097 enc_class RM_opc_mem(immI rm_opcode, memory mem)
3098 %{
3099 int rm_byte_opcode = $rm_opcode$$constant;
3101 // High registers handle in encode_RegMem
3102 int base = $mem$$base;
3103 int index = $mem$$index;
3104 int scale = $mem$$scale;
3105 int displace = $mem$$disp;
3107 bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
3108 // working with static
3109 // globals
3110 encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
3111 disp_is_oop);
3112 %}
3114 enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
3115 %{
3116 int reg_encoding = $dst$$reg;
3117 int base = $src0$$reg; // 0xFFFFFFFF indicates no base
3118 int index = 0x04; // 0x04 indicates no index
3119 int scale = 0x00; // 0x00 indicates no scale
3120 int displace = $src1$$constant; // 0x00 indicates no displacement
3121 bool disp_is_oop = false;
3122 encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
3123 disp_is_oop);
3124 %}
3126 enc_class neg_reg(rRegI dst)
3127 %{
3128 int dstenc = $dst$$reg;
3129 if (dstenc >= 8) {
3130 emit_opcode(cbuf, Assembler::REX_B);
3131 dstenc -= 8;
3132 }
3133 // NEG $dst
3134 emit_opcode(cbuf, 0xF7);
3135 emit_rm(cbuf, 0x3, 0x03, dstenc);
3136 %}
3138 enc_class neg_reg_wide(rRegI dst)
3139 %{
3140 int dstenc = $dst$$reg;
3141 if (dstenc < 8) {
3142 emit_opcode(cbuf, Assembler::REX_W);
3143 } else {
3144 emit_opcode(cbuf, Assembler::REX_WB);
3145 dstenc -= 8;
3146 }
3147 // NEG $dst
3148 emit_opcode(cbuf, 0xF7);
3149 emit_rm(cbuf, 0x3, 0x03, dstenc);
3150 %}
3152 enc_class setLT_reg(rRegI dst)
3153 %{
3154 int dstenc = $dst$$reg;
3155 if (dstenc >= 8) {
3156 emit_opcode(cbuf, Assembler::REX_B);
3157 dstenc -= 8;
3158 } else if (dstenc >= 4) {
3159 emit_opcode(cbuf, Assembler::REX);
3160 }
3161 // SETLT $dst
3162 emit_opcode(cbuf, 0x0F);
3163 emit_opcode(cbuf, 0x9C);
3164 emit_rm(cbuf, 0x3, 0x0, dstenc);
3165 %}
3167 enc_class setNZ_reg(rRegI dst)
3168 %{
3169 int dstenc = $dst$$reg;
3170 if (dstenc >= 8) {
3171 emit_opcode(cbuf, Assembler::REX_B);
3172 dstenc -= 8;
3173 } else if (dstenc >= 4) {
3174 emit_opcode(cbuf, Assembler::REX);
3175 }
3176 // SETNZ $dst
3177 emit_opcode(cbuf, 0x0F);
3178 emit_opcode(cbuf, 0x95);
3179 emit_rm(cbuf, 0x3, 0x0, dstenc);
3180 %}
3182 enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y,
3183 rcx_RegI tmp)
3184 %{
3185 // cadd_cmpLT
3187 int tmpReg = $tmp$$reg;
3189 int penc = $p$$reg;
3190 int qenc = $q$$reg;
3191 int yenc = $y$$reg;
3193 // subl $p,$q
3194 if (penc < 8) {
3195 if (qenc >= 8) {
3196 emit_opcode(cbuf, Assembler::REX_B);
3197 }
3198 } else {
3199 if (qenc < 8) {
3200 emit_opcode(cbuf, Assembler::REX_R);
3201 } else {
3202 emit_opcode(cbuf, Assembler::REX_RB);
3203 }
3204 }
3205 emit_opcode(cbuf, 0x2B);
3206 emit_rm(cbuf, 0x3, penc & 7, qenc & 7);
3208 // sbbl $tmp, $tmp
3209 emit_opcode(cbuf, 0x1B);
3210 emit_rm(cbuf, 0x3, tmpReg, tmpReg);
3212 // andl $tmp, $y
3213 if (yenc >= 8) {
3214 emit_opcode(cbuf, Assembler::REX_B);
3215 }
3216 emit_opcode(cbuf, 0x23);
3217 emit_rm(cbuf, 0x3, tmpReg, yenc & 7);
3219 // addl $p,$tmp
3220 if (penc >= 8) {
3221 emit_opcode(cbuf, Assembler::REX_R);
3222 }
3223 emit_opcode(cbuf, 0x03);
3224 emit_rm(cbuf, 0x3, penc & 7, tmpReg);
3225 %}
3227 // Compare the lonogs and set -1, 0, or 1 into dst
3228 enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
3229 %{
3230 int src1enc = $src1$$reg;
3231 int src2enc = $src2$$reg;
3232 int dstenc = $dst$$reg;
3234 // cmpq $src1, $src2
3235 if (src1enc < 8) {
3236 if (src2enc < 8) {
3237 emit_opcode(cbuf, Assembler::REX_W);
3238 } else {
3239 emit_opcode(cbuf, Assembler::REX_WB);
3240 }
3241 } else {
3242 if (src2enc < 8) {
3243 emit_opcode(cbuf, Assembler::REX_WR);
3244 } else {
3245 emit_opcode(cbuf, Assembler::REX_WRB);
3246 }
3247 }
3248 emit_opcode(cbuf, 0x3B);
3249 emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
3251 // movl $dst, -1
3252 if (dstenc >= 8) {
3253 emit_opcode(cbuf, Assembler::REX_B);
3254 }
3255 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
3256 emit_d32(cbuf, -1);
3258 // jl,s done
3259 emit_opcode(cbuf, 0x7C);
3260 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
3262 // setne $dst
3263 if (dstenc >= 4) {
3264 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
3265 }
3266 emit_opcode(cbuf, 0x0F);
3267 emit_opcode(cbuf, 0x95);
3268 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
3270 // movzbl $dst, $dst
3271 if (dstenc >= 4) {
3272 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
3273 }
3274 emit_opcode(cbuf, 0x0F);
3275 emit_opcode(cbuf, 0xB6);
3276 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
3277 %}
3279 enc_class Push_ResultXD(regD dst) %{
3280 int dstenc = $dst$$reg;
3282 store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
3284 // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
3285 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
3286 if (dstenc >= 8) {
3287 emit_opcode(cbuf, Assembler::REX_R);
3288 }
3289 emit_opcode (cbuf, 0x0F );
3290 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
3291 encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
3293 // add rsp,8
3294 emit_opcode(cbuf, Assembler::REX_W);
3295 emit_opcode(cbuf,0x83);
3296 emit_rm(cbuf,0x3, 0x0, RSP_enc);
3297 emit_d8(cbuf,0x08);
3298 %}
3300 enc_class Push_SrcXD(regD src) %{
3301 int srcenc = $src$$reg;
3303 // subq rsp,#8
3304 emit_opcode(cbuf, Assembler::REX_W);
3305 emit_opcode(cbuf, 0x83);
3306 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3307 emit_d8(cbuf, 0x8);
3309 // movsd [rsp],src
3310 emit_opcode(cbuf, 0xF2);
3311 if (srcenc >= 8) {
3312 emit_opcode(cbuf, Assembler::REX_R);
3313 }
3314 emit_opcode(cbuf, 0x0F);
3315 emit_opcode(cbuf, 0x11);
3316 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
3318 // fldd [rsp]
3319 emit_opcode(cbuf, 0x66);
3320 emit_opcode(cbuf, 0xDD);
3321 encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
3322 %}
3325 enc_class movq_ld(regD dst, memory mem) %{
3326 MacroAssembler _masm(&cbuf);
3327 __ movq($dst$$XMMRegister, $mem$$Address);
3328 %}
3330 enc_class movq_st(memory mem, regD src) %{
3331 MacroAssembler _masm(&cbuf);
3332 __ movq($mem$$Address, $src$$XMMRegister);
3333 %}
3335 enc_class pshufd_8x8(regF dst, regF src) %{
3336 MacroAssembler _masm(&cbuf);
3338 encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
3339 __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
3340 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
3341 %}
3343 enc_class pshufd_4x16(regF dst, regF src) %{
3344 MacroAssembler _masm(&cbuf);
3346 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
3347 %}
3349 enc_class pshufd(regD dst, regD src, int mode) %{
3350 MacroAssembler _masm(&cbuf);
3352 __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
3353 %}
3355 enc_class pxor(regD dst, regD src) %{
3356 MacroAssembler _masm(&cbuf);
3358 __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
3359 %}
3361 enc_class mov_i2x(regD dst, rRegI src) %{
3362 MacroAssembler _masm(&cbuf);
3364 __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
3365 %}
3367 // obj: object to lock
3368 // box: box address (header location) -- killed
3369 // tmp: rax -- killed
3370 // scr: rbx -- killed
3371 //
3372 // What follows is a direct transliteration of fast_lock() and fast_unlock()
3373 // from i486.ad. See that file for comments.
3374 // TODO: where possible switch from movq (r, 0) to movl(r,0) and
3375 // use the shorter encoding. (Movl clears the high-order 32-bits).
3378 enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
3379 %{
3380 Register objReg = as_Register((int)$obj$$reg);
3381 Register boxReg = as_Register((int)$box$$reg);
3382 Register tmpReg = as_Register($tmp$$reg);
3383 Register scrReg = as_Register($scr$$reg);
3384 MacroAssembler masm(&cbuf);
3386 // Verify uniqueness of register assignments -- necessary but not sufficient
3387 assert (objReg != boxReg && objReg != tmpReg &&
3388 objReg != scrReg && tmpReg != scrReg, "invariant") ;
3390 if (_counters != NULL) {
3391 masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3392 }
3393 if (EmitSync & 1) {
3394 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3395 masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3396 masm.cmpptr(rsp, (int32_t)NULL_WORD) ;
3397 } else
3398 if (EmitSync & 2) {
3399 Label DONE_LABEL;
3400 if (UseBiasedLocking) {
3401 // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3402 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3403 }
3404 // QQQ was movl...
3405 masm.movptr(tmpReg, 0x1);
3406 masm.orptr(tmpReg, Address(objReg, 0));
3407 masm.movptr(Address(boxReg, 0), tmpReg);
3408 if (os::is_MP()) {
3409 masm.lock();
3410 }
3411 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3412 masm.jcc(Assembler::equal, DONE_LABEL);
3414 // Recursive locking
3415 masm.subptr(tmpReg, rsp);
3416 masm.andptr(tmpReg, 7 - os::vm_page_size());
3417 masm.movptr(Address(boxReg, 0), tmpReg);
3419 masm.bind(DONE_LABEL);
3420 masm.nop(); // avoid branch to branch
3421 } else {
3422 Label DONE_LABEL, IsInflated, Egress;
3424 masm.movptr(tmpReg, Address(objReg, 0)) ;
3425 masm.testl (tmpReg, 0x02) ; // inflated vs stack-locked|neutral|biased
3426 masm.jcc (Assembler::notZero, IsInflated) ;
3428 // it's stack-locked, biased or neutral
3429 // TODO: optimize markword triage order to reduce the number of
3430 // conditional branches in the most common cases.
3431 // Beware -- there's a subtle invariant that fetch of the markword
3432 // at [FETCH], below, will never observe a biased encoding (*101b).
3433 // If this invariant is not held we'll suffer exclusion (safety) failure.
3435 if (UseBiasedLocking && !UseOptoBiasInlining) {
3436 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
3437 masm.movptr(tmpReg, Address(objReg, 0)) ; // [FETCH]
3438 }
3440 // was q will it destroy high?
3441 masm.orl (tmpReg, 1) ;
3442 masm.movptr(Address(boxReg, 0), tmpReg) ;
3443 if (os::is_MP()) { masm.lock(); }
3444 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3445 if (_counters != NULL) {
3446 masm.cond_inc32(Assembler::equal,
3447 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3448 }
3449 masm.jcc (Assembler::equal, DONE_LABEL);
3451 // Recursive locking
3452 masm.subptr(tmpReg, rsp);
3453 masm.andptr(tmpReg, 7 - os::vm_page_size());
3454 masm.movptr(Address(boxReg, 0), tmpReg);
3455 if (_counters != NULL) {
3456 masm.cond_inc32(Assembler::equal,
3457 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3458 }
3459 masm.jmp (DONE_LABEL) ;
3461 masm.bind (IsInflated) ;
3462 // It's inflated
3464 // TODO: someday avoid the ST-before-CAS penalty by
3465 // relocating (deferring) the following ST.
3466 // We should also think about trying a CAS without having
3467 // fetched _owner. If the CAS is successful we may
3468 // avoid an RTO->RTS upgrade on the $line.
3469 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3470 masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3472 masm.mov (boxReg, tmpReg) ;
3473 masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3474 masm.testptr(tmpReg, tmpReg) ;
3475 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3477 // It's inflated and appears unlocked
3478 if (os::is_MP()) { masm.lock(); }
3479 masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3480 // Intentional fall-through into DONE_LABEL ...
3482 masm.bind (DONE_LABEL) ;
3483 masm.nop () ; // avoid jmp to jmp
3484 }
3485 %}
3487 // obj: object to unlock
3488 // box: box address (displaced header location), killed
3489 // RBX: killed tmp; cannot be obj nor box
3490 enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
3491 %{
3493 Register objReg = as_Register($obj$$reg);
3494 Register boxReg = as_Register($box$$reg);
3495 Register tmpReg = as_Register($tmp$$reg);
3496 MacroAssembler masm(&cbuf);
3498 if (EmitSync & 4) {
3499 masm.cmpptr(rsp, 0) ;
3500 } else
3501 if (EmitSync & 8) {
3502 Label DONE_LABEL;
3503 if (UseBiasedLocking) {
3504 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3505 }
3507 // Check whether the displaced header is 0
3508 //(=> recursive unlock)
3509 masm.movptr(tmpReg, Address(boxReg, 0));
3510 masm.testptr(tmpReg, tmpReg);
3511 masm.jcc(Assembler::zero, DONE_LABEL);
3513 // If not recursive lock, reset the header to displaced header
3514 if (os::is_MP()) {
3515 masm.lock();
3516 }
3517 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3518 masm.bind(DONE_LABEL);
3519 masm.nop(); // avoid branch to branch
3520 } else {
3521 Label DONE_LABEL, Stacked, CheckSucc ;
3523 if (UseBiasedLocking && !UseOptoBiasInlining) {
3524 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3525 }
3527 masm.movptr(tmpReg, Address(objReg, 0)) ;
3528 masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ;
3529 masm.jcc (Assembler::zero, DONE_LABEL) ;
3530 masm.testl (tmpReg, 0x02) ;
3531 masm.jcc (Assembler::zero, Stacked) ;
3533 // It's inflated
3534 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3535 masm.xorptr(boxReg, r15_thread) ;
3536 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3537 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3538 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ;
3539 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ;
3540 masm.jcc (Assembler::notZero, CheckSucc) ;
3541 masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3542 masm.jmp (DONE_LABEL) ;
3544 if ((EmitSync & 65536) == 0) {
3545 Label LSuccess, LGoSlowPath ;
3546 masm.bind (CheckSucc) ;
3547 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3548 masm.jcc (Assembler::zero, LGoSlowPath) ;
3550 // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
3551 // the explicit ST;MEMBAR combination, but masm doesn't currently support
3552 // "ANDQ M,IMM". Don't use MFENCE here. lock:add to TOS, xchg, etc
3553 // are all faster when the write buffer is populated.
3554 masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3555 if (os::is_MP()) {
3556 masm.lock () ; masm.addl (Address(rsp, 0), 0) ;
3557 }
3558 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3559 masm.jcc (Assembler::notZero, LSuccess) ;
3561 masm.movptr (boxReg, (int32_t)NULL_WORD) ; // box is really EAX
3562 if (os::is_MP()) { masm.lock(); }
3563 masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3564 masm.jcc (Assembler::notEqual, LSuccess) ;
3565 // Intentional fall-through into slow-path
3567 masm.bind (LGoSlowPath) ;
3568 masm.orl (boxReg, 1) ; // set ICC.ZF=0 to indicate failure
3569 masm.jmp (DONE_LABEL) ;
3571 masm.bind (LSuccess) ;
3572 masm.testl (boxReg, 0) ; // set ICC.ZF=1 to indicate success
3573 masm.jmp (DONE_LABEL) ;
3574 }
3576 masm.bind (Stacked) ;
3577 masm.movptr(tmpReg, Address (boxReg, 0)) ; // re-fetch
3578 if (os::is_MP()) { masm.lock(); }
3579 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3581 if (EmitSync & 65536) {
3582 masm.bind (CheckSucc) ;
3583 }
3584 masm.bind(DONE_LABEL);
3585 if (EmitSync & 32768) {
3586 masm.nop(); // avoid branch to branch
3587 }
3588 }
3589 %}
3592 enc_class enc_rethrow()
3593 %{
3594 cbuf.set_insts_mark();
3595 emit_opcode(cbuf, 0xE9); // jmp entry
3596 emit_d32_reloc(cbuf,
3597 (int) (OptoRuntime::rethrow_stub() - cbuf.insts_end() - 4),
3598 runtime_call_Relocation::spec(),
3599 RELOC_DISP32);
3600 %}
3602 enc_class absF_encoding(regF dst)
3603 %{
3604 int dstenc = $dst$$reg;
3605 address signmask_address = (address) StubRoutines::x86::float_sign_mask();
3607 cbuf.set_insts_mark();
3608 if (dstenc >= 8) {
3609 emit_opcode(cbuf, Assembler::REX_R);
3610 dstenc -= 8;
3611 }
3612 // XXX reg_mem doesn't support RIP-relative addressing yet
3613 emit_opcode(cbuf, 0x0F);
3614 emit_opcode(cbuf, 0x54);
3615 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3616 emit_d32_reloc(cbuf, signmask_address);
3617 %}
3619 enc_class absD_encoding(regD dst)
3620 %{
3621 int dstenc = $dst$$reg;
3622 address signmask_address = (address) StubRoutines::x86::double_sign_mask();
3624 cbuf.set_insts_mark();
3625 emit_opcode(cbuf, 0x66);
3626 if (dstenc >= 8) {
3627 emit_opcode(cbuf, Assembler::REX_R);
3628 dstenc -= 8;
3629 }
3630 // XXX reg_mem doesn't support RIP-relative addressing yet
3631 emit_opcode(cbuf, 0x0F);
3632 emit_opcode(cbuf, 0x54);
3633 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3634 emit_d32_reloc(cbuf, signmask_address);
3635 %}
3637 enc_class negF_encoding(regF dst)
3638 %{
3639 int dstenc = $dst$$reg;
3640 address signflip_address = (address) StubRoutines::x86::float_sign_flip();
3642 cbuf.set_insts_mark();
3643 if (dstenc >= 8) {
3644 emit_opcode(cbuf, Assembler::REX_R);
3645 dstenc -= 8;
3646 }
3647 // XXX reg_mem doesn't support RIP-relative addressing yet
3648 emit_opcode(cbuf, 0x0F);
3649 emit_opcode(cbuf, 0x57);
3650 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3651 emit_d32_reloc(cbuf, signflip_address);
3652 %}
3654 enc_class negD_encoding(regD dst)
3655 %{
3656 int dstenc = $dst$$reg;
3657 address signflip_address = (address) StubRoutines::x86::double_sign_flip();
3659 cbuf.set_insts_mark();
3660 emit_opcode(cbuf, 0x66);
3661 if (dstenc >= 8) {
3662 emit_opcode(cbuf, Assembler::REX_R);
3663 dstenc -= 8;
3664 }
3665 // XXX reg_mem doesn't support RIP-relative addressing yet
3666 emit_opcode(cbuf, 0x0F);
3667 emit_opcode(cbuf, 0x57);
3668 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3669 emit_d32_reloc(cbuf, signflip_address);
3670 %}
3672 enc_class f2i_fixup(rRegI dst, regF src)
3673 %{
3674 int dstenc = $dst$$reg;
3675 int srcenc = $src$$reg;
3677 // cmpl $dst, #0x80000000
3678 if (dstenc >= 8) {
3679 emit_opcode(cbuf, Assembler::REX_B);
3680 }
3681 emit_opcode(cbuf, 0x81);
3682 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3683 emit_d32(cbuf, 0x80000000);
3685 // jne,s done
3686 emit_opcode(cbuf, 0x75);
3687 if (srcenc < 8 && dstenc < 8) {
3688 emit_d8(cbuf, 0xF);
3689 } else if (srcenc >= 8 && dstenc >= 8) {
3690 emit_d8(cbuf, 0x11);
3691 } else {
3692 emit_d8(cbuf, 0x10);
3693 }
3695 // subq rsp, #8
3696 emit_opcode(cbuf, Assembler::REX_W);
3697 emit_opcode(cbuf, 0x83);
3698 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3699 emit_d8(cbuf, 8);
3701 // movss [rsp], $src
3702 emit_opcode(cbuf, 0xF3);
3703 if (srcenc >= 8) {
3704 emit_opcode(cbuf, Assembler::REX_R);
3705 }
3706 emit_opcode(cbuf, 0x0F);
3707 emit_opcode(cbuf, 0x11);
3708 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3710 // call f2i_fixup
3711 cbuf.set_insts_mark();
3712 emit_opcode(cbuf, 0xE8);
3713 emit_d32_reloc(cbuf,
3714 (int)
3715 (StubRoutines::x86::f2i_fixup() - cbuf.insts_end() - 4),
3716 runtime_call_Relocation::spec(),
3717 RELOC_DISP32);
3719 // popq $dst
3720 if (dstenc >= 8) {
3721 emit_opcode(cbuf, Assembler::REX_B);
3722 }
3723 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3725 // done:
3726 %}
3728 enc_class f2l_fixup(rRegL dst, regF src)
3729 %{
3730 int dstenc = $dst$$reg;
3731 int srcenc = $src$$reg;
3732 address const_address = (address) StubRoutines::x86::double_sign_flip();
3734 // cmpq $dst, [0x8000000000000000]
3735 cbuf.set_insts_mark();
3736 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
3737 emit_opcode(cbuf, 0x39);
3738 // XXX reg_mem doesn't support RIP-relative addressing yet
3739 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
3740 emit_d32_reloc(cbuf, const_address);
3743 // jne,s done
3744 emit_opcode(cbuf, 0x75);
3745 if (srcenc < 8 && dstenc < 8) {
3746 emit_d8(cbuf, 0xF);
3747 } else if (srcenc >= 8 && dstenc >= 8) {
3748 emit_d8(cbuf, 0x11);
3749 } else {
3750 emit_d8(cbuf, 0x10);
3751 }
3753 // subq rsp, #8
3754 emit_opcode(cbuf, Assembler::REX_W);
3755 emit_opcode(cbuf, 0x83);
3756 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3757 emit_d8(cbuf, 8);
3759 // movss [rsp], $src
3760 emit_opcode(cbuf, 0xF3);
3761 if (srcenc >= 8) {
3762 emit_opcode(cbuf, Assembler::REX_R);
3763 }
3764 emit_opcode(cbuf, 0x0F);
3765 emit_opcode(cbuf, 0x11);
3766 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3768 // call f2l_fixup
3769 cbuf.set_insts_mark();
3770 emit_opcode(cbuf, 0xE8);
3771 emit_d32_reloc(cbuf,
3772 (int)
3773 (StubRoutines::x86::f2l_fixup() - cbuf.insts_end() - 4),
3774 runtime_call_Relocation::spec(),
3775 RELOC_DISP32);
3777 // popq $dst
3778 if (dstenc >= 8) {
3779 emit_opcode(cbuf, Assembler::REX_B);
3780 }
3781 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3783 // done:
3784 %}
3786 enc_class d2i_fixup(rRegI dst, regD src)
3787 %{
3788 int dstenc = $dst$$reg;
3789 int srcenc = $src$$reg;
3791 // cmpl $dst, #0x80000000
3792 if (dstenc >= 8) {
3793 emit_opcode(cbuf, Assembler::REX_B);
3794 }
3795 emit_opcode(cbuf, 0x81);
3796 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3797 emit_d32(cbuf, 0x80000000);
3799 // jne,s done
3800 emit_opcode(cbuf, 0x75);
3801 if (srcenc < 8 && dstenc < 8) {
3802 emit_d8(cbuf, 0xF);
3803 } else if (srcenc >= 8 && dstenc >= 8) {
3804 emit_d8(cbuf, 0x11);
3805 } else {
3806 emit_d8(cbuf, 0x10);
3807 }
3809 // subq rsp, #8
3810 emit_opcode(cbuf, Assembler::REX_W);
3811 emit_opcode(cbuf, 0x83);
3812 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3813 emit_d8(cbuf, 8);
3815 // movsd [rsp], $src
3816 emit_opcode(cbuf, 0xF2);
3817 if (srcenc >= 8) {
3818 emit_opcode(cbuf, Assembler::REX_R);
3819 }
3820 emit_opcode(cbuf, 0x0F);
3821 emit_opcode(cbuf, 0x11);
3822 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3824 // call d2i_fixup
3825 cbuf.set_insts_mark();
3826 emit_opcode(cbuf, 0xE8);
3827 emit_d32_reloc(cbuf,
3828 (int)
3829 (StubRoutines::x86::d2i_fixup() - cbuf.insts_end() - 4),
3830 runtime_call_Relocation::spec(),
3831 RELOC_DISP32);
3833 // popq $dst
3834 if (dstenc >= 8) {
3835 emit_opcode(cbuf, Assembler::REX_B);
3836 }
3837 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3839 // done:
3840 %}
3842 enc_class d2l_fixup(rRegL dst, regD src)
3843 %{
3844 int dstenc = $dst$$reg;
3845 int srcenc = $src$$reg;
3846 address const_address = (address) StubRoutines::x86::double_sign_flip();
3848 // cmpq $dst, [0x8000000000000000]
3849 cbuf.set_insts_mark();
3850 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
3851 emit_opcode(cbuf, 0x39);
3852 // XXX reg_mem doesn't support RIP-relative addressing yet
3853 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
3854 emit_d32_reloc(cbuf, const_address);
3857 // jne,s done
3858 emit_opcode(cbuf, 0x75);
3859 if (srcenc < 8 && dstenc < 8) {
3860 emit_d8(cbuf, 0xF);
3861 } else if (srcenc >= 8 && dstenc >= 8) {
3862 emit_d8(cbuf, 0x11);
3863 } else {
3864 emit_d8(cbuf, 0x10);
3865 }
3867 // subq rsp, #8
3868 emit_opcode(cbuf, Assembler::REX_W);
3869 emit_opcode(cbuf, 0x83);
3870 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3871 emit_d8(cbuf, 8);
3873 // movsd [rsp], $src
3874 emit_opcode(cbuf, 0xF2);
3875 if (srcenc >= 8) {
3876 emit_opcode(cbuf, Assembler::REX_R);
3877 }
3878 emit_opcode(cbuf, 0x0F);
3879 emit_opcode(cbuf, 0x11);
3880 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3882 // call d2l_fixup
3883 cbuf.set_insts_mark();
3884 emit_opcode(cbuf, 0xE8);
3885 emit_d32_reloc(cbuf,
3886 (int)
3887 (StubRoutines::x86::d2l_fixup() - cbuf.insts_end() - 4),
3888 runtime_call_Relocation::spec(),
3889 RELOC_DISP32);
3891 // popq $dst
3892 if (dstenc >= 8) {
3893 emit_opcode(cbuf, Assembler::REX_B);
3894 }
3895 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3897 // done:
3898 %}
3899 %}
3903 //----------FRAME--------------------------------------------------------------
3904 // Definition of frame structure and management information.
3905 //
3906 // S T A C K L A Y O U T Allocators stack-slot number
3907 // | (to get allocators register number
3908 // G Owned by | | v add OptoReg::stack0())
3909 // r CALLER | |
3910 // o | +--------+ pad to even-align allocators stack-slot
3911 // w V | pad0 | numbers; owned by CALLER
3912 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
3913 // h ^ | in | 5
3914 // | | args | 4 Holes in incoming args owned by SELF
3915 // | | | | 3
3916 // | | +--------+
3917 // V | | old out| Empty on Intel, window on Sparc
3918 // | old |preserve| Must be even aligned.
3919 // | SP-+--------+----> Matcher::_old_SP, even aligned
3920 // | | in | 3 area for Intel ret address
3921 // Owned by |preserve| Empty on Sparc.
3922 // SELF +--------+
3923 // | | pad2 | 2 pad to align old SP
3924 // | +--------+ 1
3925 // | | locks | 0
3926 // | +--------+----> OptoReg::stack0(), even aligned
3927 // | | pad1 | 11 pad to align new SP
3928 // | +--------+
3929 // | | | 10
3930 // | | spills | 9 spills
3931 // V | | 8 (pad0 slot for callee)
3932 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
3933 // ^ | out | 7
3934 // | | args | 6 Holes in outgoing args owned by CALLEE
3935 // Owned by +--------+
3936 // CALLEE | new out| 6 Empty on Intel, window on Sparc
3937 // | new |preserve| Must be even-aligned.
3938 // | SP-+--------+----> Matcher::_new_SP, even aligned
3939 // | | |
3940 //
3941 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
3942 // known from SELF's arguments and the Java calling convention.
3943 // Region 6-7 is determined per call site.
3944 // Note 2: If the calling convention leaves holes in the incoming argument
3945 // area, those holes are owned by SELF. Holes in the outgoing area
3946 // are owned by the CALLEE. Holes should not be nessecary in the
3947 // incoming area, as the Java calling convention is completely under
3948 // the control of the AD file. Doubles can be sorted and packed to
3949 // avoid holes. Holes in the outgoing arguments may be nessecary for
3950 // varargs C calling conventions.
3951 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
3952 // even aligned with pad0 as needed.
3953 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
3954 // region 6-11 is even aligned; it may be padded out more so that
3955 // the region from SP to FP meets the minimum stack alignment.
3956 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
3957 // alignment. Region 11, pad1, may be dynamically extended so that
3958 // SP meets the minimum alignment.
3960 frame
3961 %{
3962 // What direction does stack grow in (assumed to be same for C & Java)
3963 stack_direction(TOWARDS_LOW);
3965 // These three registers define part of the calling convention
3966 // between compiled code and the interpreter.
3967 inline_cache_reg(RAX); // Inline Cache Register
3968 interpreter_method_oop_reg(RBX); // Method Oop Register when
3969 // calling interpreter
3971 // Optional: name the operand used by cisc-spilling to access
3972 // [stack_pointer + offset]
3973 cisc_spilling_operand_name(indOffset32);
3975 // Number of stack slots consumed by locking an object
3976 sync_stack_slots(2);
3978 // Compiled code's Frame Pointer
3979 frame_pointer(RSP);
3981 // Interpreter stores its frame pointer in a register which is
3982 // stored to the stack by I2CAdaptors.
3983 // I2CAdaptors convert from interpreted java to compiled java.
3984 interpreter_frame_pointer(RBP);
3986 // Stack alignment requirement
3987 stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
3989 // Number of stack slots between incoming argument block and the start of
3990 // a new frame. The PROLOG must add this many slots to the stack. The
3991 // EPILOG must remove this many slots. amd64 needs two slots for
3992 // return address.
3993 in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
3995 // Number of outgoing stack slots killed above the out_preserve_stack_slots
3996 // for calls to C. Supports the var-args backing area for register parms.
3997 varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
3999 // The after-PROLOG location of the return address. Location of
4000 // return address specifies a type (REG or STACK) and a number
4001 // representing the register number (i.e. - use a register name) or
4002 // stack slot.
4003 // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
4004 // Otherwise, it is above the locks and verification slot and alignment word
4005 return_addr(STACK - 2 +
4006 round_to(2 + 2 * VerifyStackAtCalls +
4007 Compile::current()->fixed_slots(),
4008 WordsPerLong * 2));
4010 // Body of function which returns an integer array locating
4011 // arguments either in registers or in stack slots. Passed an array
4012 // of ideal registers called "sig" and a "length" count. Stack-slot
4013 // offsets are based on outgoing arguments, i.e. a CALLER setting up
4014 // arguments for a CALLEE. Incoming stack arguments are
4015 // automatically biased by the preserve_stack_slots field above.
4017 calling_convention
4018 %{
4019 // No difference between ingoing/outgoing just pass false
4020 SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
4021 %}
4023 c_calling_convention
4024 %{
4025 // This is obviously always outgoing
4026 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
4027 %}
4029 // Location of compiled Java return values. Same as C for now.
4030 return_value
4031 %{
4032 assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
4033 "only return normal values");
4035 static const int lo[Op_RegL + 1] = {
4036 0,
4037 0,
4038 RAX_num, // Op_RegN
4039 RAX_num, // Op_RegI
4040 RAX_num, // Op_RegP
4041 XMM0_num, // Op_RegF
4042 XMM0_num, // Op_RegD
4043 RAX_num // Op_RegL
4044 };
4045 static const int hi[Op_RegL + 1] = {
4046 0,
4047 0,
4048 OptoReg::Bad, // Op_RegN
4049 OptoReg::Bad, // Op_RegI
4050 RAX_H_num, // Op_RegP
4051 OptoReg::Bad, // Op_RegF
4052 XMM0_H_num, // Op_RegD
4053 RAX_H_num // Op_RegL
4054 };
4055 assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
4056 return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
4057 %}
4058 %}
4060 //----------ATTRIBUTES---------------------------------------------------------
4061 //----------Operand Attributes-------------------------------------------------
4062 op_attrib op_cost(0); // Required cost attribute
4064 //----------Instruction Attributes---------------------------------------------
4065 ins_attrib ins_cost(100); // Required cost attribute
4066 ins_attrib ins_size(8); // Required size attribute (in bits)
4067 ins_attrib ins_pc_relative(0); // Required PC Relative flag
4068 ins_attrib ins_short_branch(0); // Required flag: is this instruction
4069 // a non-matching short branch variant
4070 // of some long branch?
4071 ins_attrib ins_alignment(1); // Required alignment attribute (must
4072 // be a power of 2) specifies the
4073 // alignment that some part of the
4074 // instruction (not necessarily the
4075 // start) requires. If > 1, a
4076 // compute_padding() function must be
4077 // provided for the instruction
4079 //----------OPERANDS-----------------------------------------------------------
4080 // Operand definitions must precede instruction definitions for correct parsing
4081 // in the ADLC because operands constitute user defined types which are used in
4082 // instruction definitions.
4084 //----------Simple Operands----------------------------------------------------
4085 // Immediate Operands
4086 // Integer Immediate
4087 operand immI()
4088 %{
4089 match(ConI);
4091 op_cost(10);
4092 format %{ %}
4093 interface(CONST_INTER);
4094 %}
4096 // Constant for test vs zero
4097 operand immI0()
4098 %{
4099 predicate(n->get_int() == 0);
4100 match(ConI);
4102 op_cost(0);
4103 format %{ %}
4104 interface(CONST_INTER);
4105 %}
4107 // Constant for increment
4108 operand immI1()
4109 %{
4110 predicate(n->get_int() == 1);
4111 match(ConI);
4113 op_cost(0);
4114 format %{ %}
4115 interface(CONST_INTER);
4116 %}
4118 // Constant for decrement
4119 operand immI_M1()
4120 %{
4121 predicate(n->get_int() == -1);
4122 match(ConI);
4124 op_cost(0);
4125 format %{ %}
4126 interface(CONST_INTER);
4127 %}
4129 // Valid scale values for addressing modes
4130 operand immI2()
4131 %{
4132 predicate(0 <= n->get_int() && (n->get_int() <= 3));
4133 match(ConI);
4135 format %{ %}
4136 interface(CONST_INTER);
4137 %}
4139 operand immI8()
4140 %{
4141 predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
4142 match(ConI);
4144 op_cost(5);
4145 format %{ %}
4146 interface(CONST_INTER);
4147 %}
4149 operand immI16()
4150 %{
4151 predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
4152 match(ConI);
4154 op_cost(10);
4155 format %{ %}
4156 interface(CONST_INTER);
4157 %}
4159 // Constant for long shifts
4160 operand immI_32()
4161 %{
4162 predicate( n->get_int() == 32 );
4163 match(ConI);
4165 op_cost(0);
4166 format %{ %}
4167 interface(CONST_INTER);
4168 %}
4170 // Constant for long shifts
4171 operand immI_64()
4172 %{
4173 predicate( n->get_int() == 64 );
4174 match(ConI);
4176 op_cost(0);
4177 format %{ %}
4178 interface(CONST_INTER);
4179 %}
4181 // Pointer Immediate
4182 operand immP()
4183 %{
4184 match(ConP);
4186 op_cost(10);
4187 format %{ %}
4188 interface(CONST_INTER);
4189 %}
4191 // NULL Pointer Immediate
4192 operand immP0()
4193 %{
4194 predicate(n->get_ptr() == 0);
4195 match(ConP);
4197 op_cost(5);
4198 format %{ %}
4199 interface(CONST_INTER);
4200 %}
4202 operand immP_poll() %{
4203 predicate(n->get_ptr() != 0 && n->get_ptr() == (intptr_t)os::get_polling_page());
4204 match(ConP);
4206 // formats are generated automatically for constants and base registers
4207 format %{ %}
4208 interface(CONST_INTER);
4209 %}
4211 // Pointer Immediate
4212 operand immN() %{
4213 match(ConN);
4215 op_cost(10);
4216 format %{ %}
4217 interface(CONST_INTER);
4218 %}
4220 // NULL Pointer Immediate
4221 operand immN0() %{
4222 predicate(n->get_narrowcon() == 0);
4223 match(ConN);
4225 op_cost(5);
4226 format %{ %}
4227 interface(CONST_INTER);
4228 %}
4230 operand immP31()
4231 %{
4232 predicate(!n->as_Type()->type()->isa_oopptr()
4233 && (n->get_ptr() >> 31) == 0);
4234 match(ConP);
4236 op_cost(5);
4237 format %{ %}
4238 interface(CONST_INTER);
4239 %}
4242 // Long Immediate
4243 operand immL()
4244 %{
4245 match(ConL);
4247 op_cost(20);
4248 format %{ %}
4249 interface(CONST_INTER);
4250 %}
4252 // Long Immediate 8-bit
4253 operand immL8()
4254 %{
4255 predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
4256 match(ConL);
4258 op_cost(5);
4259 format %{ %}
4260 interface(CONST_INTER);
4261 %}
4263 // Long Immediate 32-bit unsigned
4264 operand immUL32()
4265 %{
4266 predicate(n->get_long() == (unsigned int) (n->get_long()));
4267 match(ConL);
4269 op_cost(10);
4270 format %{ %}
4271 interface(CONST_INTER);
4272 %}
4274 // Long Immediate 32-bit signed
4275 operand immL32()
4276 %{
4277 predicate(n->get_long() == (int) (n->get_long()));
4278 match(ConL);
4280 op_cost(15);
4281 format %{ %}
4282 interface(CONST_INTER);
4283 %}
4285 // Long Immediate zero
4286 operand immL0()
4287 %{
4288 predicate(n->get_long() == 0L);
4289 match(ConL);
4291 op_cost(10);
4292 format %{ %}
4293 interface(CONST_INTER);
4294 %}
4296 // Constant for increment
4297 operand immL1()
4298 %{
4299 predicate(n->get_long() == 1);
4300 match(ConL);
4302 format %{ %}
4303 interface(CONST_INTER);
4304 %}
4306 // Constant for decrement
4307 operand immL_M1()
4308 %{
4309 predicate(n->get_long() == -1);
4310 match(ConL);
4312 format %{ %}
4313 interface(CONST_INTER);
4314 %}
4316 // Long Immediate: the value 10
4317 operand immL10()
4318 %{
4319 predicate(n->get_long() == 10);
4320 match(ConL);
4322 format %{ %}
4323 interface(CONST_INTER);
4324 %}
4326 // Long immediate from 0 to 127.
4327 // Used for a shorter form of long mul by 10.
4328 operand immL_127()
4329 %{
4330 predicate(0 <= n->get_long() && n->get_long() < 0x80);
4331 match(ConL);
4333 op_cost(10);
4334 format %{ %}
4335 interface(CONST_INTER);
4336 %}
4338 // Long Immediate: low 32-bit mask
4339 operand immL_32bits()
4340 %{
4341 predicate(n->get_long() == 0xFFFFFFFFL);
4342 match(ConL);
4343 op_cost(20);
4345 format %{ %}
4346 interface(CONST_INTER);
4347 %}
4349 // Float Immediate zero
4350 operand immF0()
4351 %{
4352 predicate(jint_cast(n->getf()) == 0);
4353 match(ConF);
4355 op_cost(5);
4356 format %{ %}
4357 interface(CONST_INTER);
4358 %}
4360 // Float Immediate
4361 operand immF()
4362 %{
4363 match(ConF);
4365 op_cost(15);
4366 format %{ %}
4367 interface(CONST_INTER);
4368 %}
4370 // Double Immediate zero
4371 operand immD0()
4372 %{
4373 predicate(jlong_cast(n->getd()) == 0);
4374 match(ConD);
4376 op_cost(5);
4377 format %{ %}
4378 interface(CONST_INTER);
4379 %}
4381 // Double Immediate
4382 operand immD()
4383 %{
4384 match(ConD);
4386 op_cost(15);
4387 format %{ %}
4388 interface(CONST_INTER);
4389 %}
4391 // Immediates for special shifts (sign extend)
4393 // Constants for increment
4394 operand immI_16()
4395 %{
4396 predicate(n->get_int() == 16);
4397 match(ConI);
4399 format %{ %}
4400 interface(CONST_INTER);
4401 %}
4403 operand immI_24()
4404 %{
4405 predicate(n->get_int() == 24);
4406 match(ConI);
4408 format %{ %}
4409 interface(CONST_INTER);
4410 %}
4412 // Constant for byte-wide masking
4413 operand immI_255()
4414 %{
4415 predicate(n->get_int() == 255);
4416 match(ConI);
4418 format %{ %}
4419 interface(CONST_INTER);
4420 %}
4422 // Constant for short-wide masking
4423 operand immI_65535()
4424 %{
4425 predicate(n->get_int() == 65535);
4426 match(ConI);
4428 format %{ %}
4429 interface(CONST_INTER);
4430 %}
4432 // Constant for byte-wide masking
4433 operand immL_255()
4434 %{
4435 predicate(n->get_long() == 255);
4436 match(ConL);
4438 format %{ %}
4439 interface(CONST_INTER);
4440 %}
4442 // Constant for short-wide masking
4443 operand immL_65535()
4444 %{
4445 predicate(n->get_long() == 65535);
4446 match(ConL);
4448 format %{ %}
4449 interface(CONST_INTER);
4450 %}
4452 // Register Operands
4453 // Integer Register
4454 operand rRegI()
4455 %{
4456 constraint(ALLOC_IN_RC(int_reg));
4457 match(RegI);
4459 match(rax_RegI);
4460 match(rbx_RegI);
4461 match(rcx_RegI);
4462 match(rdx_RegI);
4463 match(rdi_RegI);
4465 format %{ %}
4466 interface(REG_INTER);
4467 %}
4469 // Special Registers
4470 operand rax_RegI()
4471 %{
4472 constraint(ALLOC_IN_RC(int_rax_reg));
4473 match(RegI);
4474 match(rRegI);
4476 format %{ "RAX" %}
4477 interface(REG_INTER);
4478 %}
4480 // Special Registers
4481 operand rbx_RegI()
4482 %{
4483 constraint(ALLOC_IN_RC(int_rbx_reg));
4484 match(RegI);
4485 match(rRegI);
4487 format %{ "RBX" %}
4488 interface(REG_INTER);
4489 %}
4491 operand rcx_RegI()
4492 %{
4493 constraint(ALLOC_IN_RC(int_rcx_reg));
4494 match(RegI);
4495 match(rRegI);
4497 format %{ "RCX" %}
4498 interface(REG_INTER);
4499 %}
4501 operand rdx_RegI()
4502 %{
4503 constraint(ALLOC_IN_RC(int_rdx_reg));
4504 match(RegI);
4505 match(rRegI);
4507 format %{ "RDX" %}
4508 interface(REG_INTER);
4509 %}
4511 operand rdi_RegI()
4512 %{
4513 constraint(ALLOC_IN_RC(int_rdi_reg));
4514 match(RegI);
4515 match(rRegI);
4517 format %{ "RDI" %}
4518 interface(REG_INTER);
4519 %}
4521 operand no_rcx_RegI()
4522 %{
4523 constraint(ALLOC_IN_RC(int_no_rcx_reg));
4524 match(RegI);
4525 match(rax_RegI);
4526 match(rbx_RegI);
4527 match(rdx_RegI);
4528 match(rdi_RegI);
4530 format %{ %}
4531 interface(REG_INTER);
4532 %}
4534 operand no_rax_rdx_RegI()
4535 %{
4536 constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
4537 match(RegI);
4538 match(rbx_RegI);
4539 match(rcx_RegI);
4540 match(rdi_RegI);
4542 format %{ %}
4543 interface(REG_INTER);
4544 %}
4546 // Pointer Register
4547 operand any_RegP()
4548 %{
4549 constraint(ALLOC_IN_RC(any_reg));
4550 match(RegP);
4551 match(rax_RegP);
4552 match(rbx_RegP);
4553 match(rdi_RegP);
4554 match(rsi_RegP);
4555 match(rbp_RegP);
4556 match(r15_RegP);
4557 match(rRegP);
4559 format %{ %}
4560 interface(REG_INTER);
4561 %}
4563 operand rRegP()
4564 %{
4565 constraint(ALLOC_IN_RC(ptr_reg));
4566 match(RegP);
4567 match(rax_RegP);
4568 match(rbx_RegP);
4569 match(rdi_RegP);
4570 match(rsi_RegP);
4571 match(rbp_RegP);
4572 match(r15_RegP); // See Q&A below about r15_RegP.
4574 format %{ %}
4575 interface(REG_INTER);
4576 %}
4578 operand rRegN() %{
4579 constraint(ALLOC_IN_RC(int_reg));
4580 match(RegN);
4582 format %{ %}
4583 interface(REG_INTER);
4584 %}
4586 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
4587 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
4588 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
4589 // The output of an instruction is controlled by the allocator, which respects
4590 // register class masks, not match rules. Unless an instruction mentions
4591 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
4592 // by the allocator as an input.
4594 operand no_rax_RegP()
4595 %{
4596 constraint(ALLOC_IN_RC(ptr_no_rax_reg));
4597 match(RegP);
4598 match(rbx_RegP);
4599 match(rsi_RegP);
4600 match(rdi_RegP);
4602 format %{ %}
4603 interface(REG_INTER);
4604 %}
4606 operand no_rbp_RegP()
4607 %{
4608 constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
4609 match(RegP);
4610 match(rbx_RegP);
4611 match(rsi_RegP);
4612 match(rdi_RegP);
4614 format %{ %}
4615 interface(REG_INTER);
4616 %}
4618 operand no_rax_rbx_RegP()
4619 %{
4620 constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
4621 match(RegP);
4622 match(rsi_RegP);
4623 match(rdi_RegP);
4625 format %{ %}
4626 interface(REG_INTER);
4627 %}
4629 // Special Registers
4630 // Return a pointer value
4631 operand rax_RegP()
4632 %{
4633 constraint(ALLOC_IN_RC(ptr_rax_reg));
4634 match(RegP);
4635 match(rRegP);
4637 format %{ %}
4638 interface(REG_INTER);
4639 %}
4641 // Special Registers
4642 // Return a compressed pointer value
4643 operand rax_RegN()
4644 %{
4645 constraint(ALLOC_IN_RC(int_rax_reg));
4646 match(RegN);
4647 match(rRegN);
4649 format %{ %}
4650 interface(REG_INTER);
4651 %}
4653 // Used in AtomicAdd
4654 operand rbx_RegP()
4655 %{
4656 constraint(ALLOC_IN_RC(ptr_rbx_reg));
4657 match(RegP);
4658 match(rRegP);
4660 format %{ %}
4661 interface(REG_INTER);
4662 %}
4664 operand rsi_RegP()
4665 %{
4666 constraint(ALLOC_IN_RC(ptr_rsi_reg));
4667 match(RegP);
4668 match(rRegP);
4670 format %{ %}
4671 interface(REG_INTER);
4672 %}
4674 // Used in rep stosq
4675 operand rdi_RegP()
4676 %{
4677 constraint(ALLOC_IN_RC(ptr_rdi_reg));
4678 match(RegP);
4679 match(rRegP);
4681 format %{ %}
4682 interface(REG_INTER);
4683 %}
4685 operand rbp_RegP()
4686 %{
4687 constraint(ALLOC_IN_RC(ptr_rbp_reg));
4688 match(RegP);
4689 match(rRegP);
4691 format %{ %}
4692 interface(REG_INTER);
4693 %}
4695 operand r15_RegP()
4696 %{
4697 constraint(ALLOC_IN_RC(ptr_r15_reg));
4698 match(RegP);
4699 match(rRegP);
4701 format %{ %}
4702 interface(REG_INTER);
4703 %}
4705 operand rRegL()
4706 %{
4707 constraint(ALLOC_IN_RC(long_reg));
4708 match(RegL);
4709 match(rax_RegL);
4710 match(rdx_RegL);
4712 format %{ %}
4713 interface(REG_INTER);
4714 %}
4716 // Special Registers
4717 operand no_rax_rdx_RegL()
4718 %{
4719 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4720 match(RegL);
4721 match(rRegL);
4723 format %{ %}
4724 interface(REG_INTER);
4725 %}
4727 operand no_rax_RegL()
4728 %{
4729 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4730 match(RegL);
4731 match(rRegL);
4732 match(rdx_RegL);
4734 format %{ %}
4735 interface(REG_INTER);
4736 %}
4738 operand no_rcx_RegL()
4739 %{
4740 constraint(ALLOC_IN_RC(long_no_rcx_reg));
4741 match(RegL);
4742 match(rRegL);
4744 format %{ %}
4745 interface(REG_INTER);
4746 %}
4748 operand rax_RegL()
4749 %{
4750 constraint(ALLOC_IN_RC(long_rax_reg));
4751 match(RegL);
4752 match(rRegL);
4754 format %{ "RAX" %}
4755 interface(REG_INTER);
4756 %}
4758 operand rcx_RegL()
4759 %{
4760 constraint(ALLOC_IN_RC(long_rcx_reg));
4761 match(RegL);
4762 match(rRegL);
4764 format %{ %}
4765 interface(REG_INTER);
4766 %}
4768 operand rdx_RegL()
4769 %{
4770 constraint(ALLOC_IN_RC(long_rdx_reg));
4771 match(RegL);
4772 match(rRegL);
4774 format %{ %}
4775 interface(REG_INTER);
4776 %}
4778 // Flags register, used as output of compare instructions
4779 operand rFlagsReg()
4780 %{
4781 constraint(ALLOC_IN_RC(int_flags));
4782 match(RegFlags);
4784 format %{ "RFLAGS" %}
4785 interface(REG_INTER);
4786 %}
4788 // Flags register, used as output of FLOATING POINT compare instructions
4789 operand rFlagsRegU()
4790 %{
4791 constraint(ALLOC_IN_RC(int_flags));
4792 match(RegFlags);
4794 format %{ "RFLAGS_U" %}
4795 interface(REG_INTER);
4796 %}
4798 operand rFlagsRegUCF() %{
4799 constraint(ALLOC_IN_RC(int_flags));
4800 match(RegFlags);
4801 predicate(false);
4803 format %{ "RFLAGS_U_CF" %}
4804 interface(REG_INTER);
4805 %}
4807 // Float register operands
4808 operand regF()
4809 %{
4810 constraint(ALLOC_IN_RC(float_reg));
4811 match(RegF);
4813 format %{ %}
4814 interface(REG_INTER);
4815 %}
4817 // Double register operands
4818 operand regD()
4819 %{
4820 constraint(ALLOC_IN_RC(double_reg));
4821 match(RegD);
4823 format %{ %}
4824 interface(REG_INTER);
4825 %}
4828 //----------Memory Operands----------------------------------------------------
4829 // Direct Memory Operand
4830 // operand direct(immP addr)
4831 // %{
4832 // match(addr);
4834 // format %{ "[$addr]" %}
4835 // interface(MEMORY_INTER) %{
4836 // base(0xFFFFFFFF);
4837 // index(0x4);
4838 // scale(0x0);
4839 // disp($addr);
4840 // %}
4841 // %}
4843 // Indirect Memory Operand
4844 operand indirect(any_RegP reg)
4845 %{
4846 constraint(ALLOC_IN_RC(ptr_reg));
4847 match(reg);
4849 format %{ "[$reg]" %}
4850 interface(MEMORY_INTER) %{
4851 base($reg);
4852 index(0x4);
4853 scale(0x0);
4854 disp(0x0);
4855 %}
4856 %}
4858 // Indirect Memory Plus Short Offset Operand
4859 operand indOffset8(any_RegP reg, immL8 off)
4860 %{
4861 constraint(ALLOC_IN_RC(ptr_reg));
4862 match(AddP reg off);
4864 format %{ "[$reg + $off (8-bit)]" %}
4865 interface(MEMORY_INTER) %{
4866 base($reg);
4867 index(0x4);
4868 scale(0x0);
4869 disp($off);
4870 %}
4871 %}
4873 // Indirect Memory Plus Long Offset Operand
4874 operand indOffset32(any_RegP reg, immL32 off)
4875 %{
4876 constraint(ALLOC_IN_RC(ptr_reg));
4877 match(AddP reg off);
4879 format %{ "[$reg + $off (32-bit)]" %}
4880 interface(MEMORY_INTER) %{
4881 base($reg);
4882 index(0x4);
4883 scale(0x0);
4884 disp($off);
4885 %}
4886 %}
4888 // Indirect Memory Plus Index Register Plus Offset Operand
4889 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
4890 %{
4891 constraint(ALLOC_IN_RC(ptr_reg));
4892 match(AddP (AddP reg lreg) off);
4894 op_cost(10);
4895 format %{"[$reg + $off + $lreg]" %}
4896 interface(MEMORY_INTER) %{
4897 base($reg);
4898 index($lreg);
4899 scale(0x0);
4900 disp($off);
4901 %}
4902 %}
4904 // Indirect Memory Plus Index Register Plus Offset Operand
4905 operand indIndex(any_RegP reg, rRegL lreg)
4906 %{
4907 constraint(ALLOC_IN_RC(ptr_reg));
4908 match(AddP reg lreg);
4910 op_cost(10);
4911 format %{"[$reg + $lreg]" %}
4912 interface(MEMORY_INTER) %{
4913 base($reg);
4914 index($lreg);
4915 scale(0x0);
4916 disp(0x0);
4917 %}
4918 %}
4920 // Indirect Memory Times Scale Plus Index Register
4921 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
4922 %{
4923 constraint(ALLOC_IN_RC(ptr_reg));
4924 match(AddP reg (LShiftL lreg scale));
4926 op_cost(10);
4927 format %{"[$reg + $lreg << $scale]" %}
4928 interface(MEMORY_INTER) %{
4929 base($reg);
4930 index($lreg);
4931 scale($scale);
4932 disp(0x0);
4933 %}
4934 %}
4936 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
4937 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
4938 %{
4939 constraint(ALLOC_IN_RC(ptr_reg));
4940 match(AddP (AddP reg (LShiftL lreg scale)) off);
4942 op_cost(10);
4943 format %{"[$reg + $off + $lreg << $scale]" %}
4944 interface(MEMORY_INTER) %{
4945 base($reg);
4946 index($lreg);
4947 scale($scale);
4948 disp($off);
4949 %}
4950 %}
4952 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
4953 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
4954 %{
4955 constraint(ALLOC_IN_RC(ptr_reg));
4956 predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
4957 match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
4959 op_cost(10);
4960 format %{"[$reg + $off + $idx << $scale]" %}
4961 interface(MEMORY_INTER) %{
4962 base($reg);
4963 index($idx);
4964 scale($scale);
4965 disp($off);
4966 %}
4967 %}
4969 // Indirect Narrow Oop Plus Offset Operand
4970 // Note: x86 architecture doesn't support "scale * index + offset" without a base
4971 // we can't free r12 even with Universe::narrow_oop_base() == NULL.
4972 operand indCompressedOopOffset(rRegN reg, immL32 off) %{
4973 predicate(UseCompressedOops && (Universe::narrow_oop_shift() == Address::times_8));
4974 constraint(ALLOC_IN_RC(ptr_reg));
4975 match(AddP (DecodeN reg) off);
4977 op_cost(10);
4978 format %{"[R12 + $reg << 3 + $off] (compressed oop addressing)" %}
4979 interface(MEMORY_INTER) %{
4980 base(0xc); // R12
4981 index($reg);
4982 scale(0x3);
4983 disp($off);
4984 %}
4985 %}
4987 // Indirect Memory Operand
4988 operand indirectNarrow(rRegN reg)
4989 %{
4990 predicate(Universe::narrow_oop_shift() == 0);
4991 constraint(ALLOC_IN_RC(ptr_reg));
4992 match(DecodeN reg);
4994 format %{ "[$reg]" %}
4995 interface(MEMORY_INTER) %{
4996 base($reg);
4997 index(0x4);
4998 scale(0x0);
4999 disp(0x0);
5000 %}
5001 %}
5003 // Indirect Memory Plus Short Offset Operand
5004 operand indOffset8Narrow(rRegN reg, immL8 off)
5005 %{
5006 predicate(Universe::narrow_oop_shift() == 0);
5007 constraint(ALLOC_IN_RC(ptr_reg));
5008 match(AddP (DecodeN reg) off);
5010 format %{ "[$reg + $off (8-bit)]" %}
5011 interface(MEMORY_INTER) %{
5012 base($reg);
5013 index(0x4);
5014 scale(0x0);
5015 disp($off);
5016 %}
5017 %}
5019 // Indirect Memory Plus Long Offset Operand
5020 operand indOffset32Narrow(rRegN reg, immL32 off)
5021 %{
5022 predicate(Universe::narrow_oop_shift() == 0);
5023 constraint(ALLOC_IN_RC(ptr_reg));
5024 match(AddP (DecodeN reg) off);
5026 format %{ "[$reg + $off (32-bit)]" %}
5027 interface(MEMORY_INTER) %{
5028 base($reg);
5029 index(0x4);
5030 scale(0x0);
5031 disp($off);
5032 %}
5033 %}
5035 // Indirect Memory Plus Index Register Plus Offset Operand
5036 operand indIndexOffsetNarrow(rRegN reg, rRegL lreg, immL32 off)
5037 %{
5038 predicate(Universe::narrow_oop_shift() == 0);
5039 constraint(ALLOC_IN_RC(ptr_reg));
5040 match(AddP (AddP (DecodeN reg) lreg) off);
5042 op_cost(10);
5043 format %{"[$reg + $off + $lreg]" %}
5044 interface(MEMORY_INTER) %{
5045 base($reg);
5046 index($lreg);
5047 scale(0x0);
5048 disp($off);
5049 %}
5050 %}
5052 // Indirect Memory Plus Index Register Plus Offset Operand
5053 operand indIndexNarrow(rRegN reg, rRegL lreg)
5054 %{
5055 predicate(Universe::narrow_oop_shift() == 0);
5056 constraint(ALLOC_IN_RC(ptr_reg));
5057 match(AddP (DecodeN reg) lreg);
5059 op_cost(10);
5060 format %{"[$reg + $lreg]" %}
5061 interface(MEMORY_INTER) %{
5062 base($reg);
5063 index($lreg);
5064 scale(0x0);
5065 disp(0x0);
5066 %}
5067 %}
5069 // Indirect Memory Times Scale Plus Index Register
5070 operand indIndexScaleNarrow(rRegN reg, rRegL lreg, immI2 scale)
5071 %{
5072 predicate(Universe::narrow_oop_shift() == 0);
5073 constraint(ALLOC_IN_RC(ptr_reg));
5074 match(AddP (DecodeN reg) (LShiftL lreg scale));
5076 op_cost(10);
5077 format %{"[$reg + $lreg << $scale]" %}
5078 interface(MEMORY_INTER) %{
5079 base($reg);
5080 index($lreg);
5081 scale($scale);
5082 disp(0x0);
5083 %}
5084 %}
5086 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5087 operand indIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegL lreg, immI2 scale)
5088 %{
5089 predicate(Universe::narrow_oop_shift() == 0);
5090 constraint(ALLOC_IN_RC(ptr_reg));
5091 match(AddP (AddP (DecodeN reg) (LShiftL lreg scale)) off);
5093 op_cost(10);
5094 format %{"[$reg + $off + $lreg << $scale]" %}
5095 interface(MEMORY_INTER) %{
5096 base($reg);
5097 index($lreg);
5098 scale($scale);
5099 disp($off);
5100 %}
5101 %}
5103 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5104 operand indPosIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegI idx, immI2 scale)
5105 %{
5106 constraint(ALLOC_IN_RC(ptr_reg));
5107 predicate(Universe::narrow_oop_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5108 match(AddP (AddP (DecodeN reg) (LShiftL (ConvI2L idx) scale)) off);
5110 op_cost(10);
5111 format %{"[$reg + $off + $idx << $scale]" %}
5112 interface(MEMORY_INTER) %{
5113 base($reg);
5114 index($idx);
5115 scale($scale);
5116 disp($off);
5117 %}
5118 %}
5121 //----------Special Memory Operands--------------------------------------------
5122 // Stack Slot Operand - This operand is used for loading and storing temporary
5123 // values on the stack where a match requires a value to
5124 // flow through memory.
5125 operand stackSlotP(sRegP reg)
5126 %{
5127 constraint(ALLOC_IN_RC(stack_slots));
5128 // No match rule because this operand is only generated in matching
5130 format %{ "[$reg]" %}
5131 interface(MEMORY_INTER) %{
5132 base(0x4); // RSP
5133 index(0x4); // No Index
5134 scale(0x0); // No Scale
5135 disp($reg); // Stack Offset
5136 %}
5137 %}
5139 operand stackSlotI(sRegI reg)
5140 %{
5141 constraint(ALLOC_IN_RC(stack_slots));
5142 // No match rule because this operand is only generated in matching
5144 format %{ "[$reg]" %}
5145 interface(MEMORY_INTER) %{
5146 base(0x4); // RSP
5147 index(0x4); // No Index
5148 scale(0x0); // No Scale
5149 disp($reg); // Stack Offset
5150 %}
5151 %}
5153 operand stackSlotF(sRegF reg)
5154 %{
5155 constraint(ALLOC_IN_RC(stack_slots));
5156 // No match rule because this operand is only generated in matching
5158 format %{ "[$reg]" %}
5159 interface(MEMORY_INTER) %{
5160 base(0x4); // RSP
5161 index(0x4); // No Index
5162 scale(0x0); // No Scale
5163 disp($reg); // Stack Offset
5164 %}
5165 %}
5167 operand stackSlotD(sRegD reg)
5168 %{
5169 constraint(ALLOC_IN_RC(stack_slots));
5170 // No match rule because this operand is only generated in matching
5172 format %{ "[$reg]" %}
5173 interface(MEMORY_INTER) %{
5174 base(0x4); // RSP
5175 index(0x4); // No Index
5176 scale(0x0); // No Scale
5177 disp($reg); // Stack Offset
5178 %}
5179 %}
5180 operand stackSlotL(sRegL reg)
5181 %{
5182 constraint(ALLOC_IN_RC(stack_slots));
5183 // No match rule because this operand is only generated in matching
5185 format %{ "[$reg]" %}
5186 interface(MEMORY_INTER) %{
5187 base(0x4); // RSP
5188 index(0x4); // No Index
5189 scale(0x0); // No Scale
5190 disp($reg); // Stack Offset
5191 %}
5192 %}
5194 //----------Conditional Branch Operands----------------------------------------
5195 // Comparison Op - This is the operation of the comparison, and is limited to
5196 // the following set of codes:
5197 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
5198 //
5199 // Other attributes of the comparison, such as unsignedness, are specified
5200 // by the comparison instruction that sets a condition code flags register.
5201 // That result is represented by a flags operand whose subtype is appropriate
5202 // to the unsignedness (etc.) of the comparison.
5203 //
5204 // Later, the instruction which matches both the Comparison Op (a Bool) and
5205 // the flags (produced by the Cmp) specifies the coding of the comparison op
5206 // by matching a specific subtype of Bool operand below, such as cmpOpU.
5208 // Comparision Code
5209 operand cmpOp()
5210 %{
5211 match(Bool);
5213 format %{ "" %}
5214 interface(COND_INTER) %{
5215 equal(0x4, "e");
5216 not_equal(0x5, "ne");
5217 less(0xC, "l");
5218 greater_equal(0xD, "ge");
5219 less_equal(0xE, "le");
5220 greater(0xF, "g");
5221 %}
5222 %}
5224 // Comparison Code, unsigned compare. Used by FP also, with
5225 // C2 (unordered) turned into GT or LT already. The other bits
5226 // C0 and C3 are turned into Carry & Zero flags.
5227 operand cmpOpU()
5228 %{
5229 match(Bool);
5231 format %{ "" %}
5232 interface(COND_INTER) %{
5233 equal(0x4, "e");
5234 not_equal(0x5, "ne");
5235 less(0x2, "b");
5236 greater_equal(0x3, "nb");
5237 less_equal(0x6, "be");
5238 greater(0x7, "nbe");
5239 %}
5240 %}
5243 // Floating comparisons that don't require any fixup for the unordered case
5244 operand cmpOpUCF() %{
5245 match(Bool);
5246 predicate(n->as_Bool()->_test._test == BoolTest::lt ||
5247 n->as_Bool()->_test._test == BoolTest::ge ||
5248 n->as_Bool()->_test._test == BoolTest::le ||
5249 n->as_Bool()->_test._test == BoolTest::gt);
5250 format %{ "" %}
5251 interface(COND_INTER) %{
5252 equal(0x4, "e");
5253 not_equal(0x5, "ne");
5254 less(0x2, "b");
5255 greater_equal(0x3, "nb");
5256 less_equal(0x6, "be");
5257 greater(0x7, "nbe");
5258 %}
5259 %}
5262 // Floating comparisons that can be fixed up with extra conditional jumps
5263 operand cmpOpUCF2() %{
5264 match(Bool);
5265 predicate(n->as_Bool()->_test._test == BoolTest::ne ||
5266 n->as_Bool()->_test._test == BoolTest::eq);
5267 format %{ "" %}
5268 interface(COND_INTER) %{
5269 equal(0x4, "e");
5270 not_equal(0x5, "ne");
5271 less(0x2, "b");
5272 greater_equal(0x3, "nb");
5273 less_equal(0x6, "be");
5274 greater(0x7, "nbe");
5275 %}
5276 %}
5279 //----------OPERAND CLASSES----------------------------------------------------
5280 // Operand Classes are groups of operands that are used as to simplify
5281 // instruction definitions by not requiring the AD writer to specify separate
5282 // instructions for every form of operand when the instruction accepts
5283 // multiple operand types with the same basic encoding and format. The classic
5284 // case of this is memory operands.
5286 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
5287 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
5288 indCompressedOopOffset,
5289 indirectNarrow, indOffset8Narrow, indOffset32Narrow,
5290 indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow,
5291 indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow);
5293 //----------PIPELINE-----------------------------------------------------------
5294 // Rules which define the behavior of the target architectures pipeline.
5295 pipeline %{
5297 //----------ATTRIBUTES---------------------------------------------------------
5298 attributes %{
5299 variable_size_instructions; // Fixed size instructions
5300 max_instructions_per_bundle = 3; // Up to 3 instructions per bundle
5301 instruction_unit_size = 1; // An instruction is 1 bytes long
5302 instruction_fetch_unit_size = 16; // The processor fetches one line
5303 instruction_fetch_units = 1; // of 16 bytes
5305 // List of nop instructions
5306 nops( MachNop );
5307 %}
5309 //----------RESOURCES----------------------------------------------------------
5310 // Resources are the functional units available to the machine
5312 // Generic P2/P3 pipeline
5313 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5314 // 3 instructions decoded per cycle.
5315 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5316 // 3 ALU op, only ALU0 handles mul instructions.
5317 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5318 MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
5319 BR, FPU,
5320 ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
5322 //----------PIPELINE DESCRIPTION-----------------------------------------------
5323 // Pipeline Description specifies the stages in the machine's pipeline
5325 // Generic P2/P3 pipeline
5326 pipe_desc(S0, S1, S2, S3, S4, S5);
5328 //----------PIPELINE CLASSES---------------------------------------------------
5329 // Pipeline Classes describe the stages in which input and output are
5330 // referenced by the hardware pipeline.
5332 // Naming convention: ialu or fpu
5333 // Then: _reg
5334 // Then: _reg if there is a 2nd register
5335 // Then: _long if it's a pair of instructions implementing a long
5336 // Then: _fat if it requires the big decoder
5337 // Or: _mem if it requires the big decoder and a memory unit.
5339 // Integer ALU reg operation
5340 pipe_class ialu_reg(rRegI dst)
5341 %{
5342 single_instruction;
5343 dst : S4(write);
5344 dst : S3(read);
5345 DECODE : S0; // any decoder
5346 ALU : S3; // any alu
5347 %}
5349 // Long ALU reg operation
5350 pipe_class ialu_reg_long(rRegL dst)
5351 %{
5352 instruction_count(2);
5353 dst : S4(write);
5354 dst : S3(read);
5355 DECODE : S0(2); // any 2 decoders
5356 ALU : S3(2); // both alus
5357 %}
5359 // Integer ALU reg operation using big decoder
5360 pipe_class ialu_reg_fat(rRegI dst)
5361 %{
5362 single_instruction;
5363 dst : S4(write);
5364 dst : S3(read);
5365 D0 : S0; // big decoder only
5366 ALU : S3; // any alu
5367 %}
5369 // Long ALU reg operation using big decoder
5370 pipe_class ialu_reg_long_fat(rRegL dst)
5371 %{
5372 instruction_count(2);
5373 dst : S4(write);
5374 dst : S3(read);
5375 D0 : S0(2); // big decoder only; twice
5376 ALU : S3(2); // any 2 alus
5377 %}
5379 // Integer ALU reg-reg operation
5380 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
5381 %{
5382 single_instruction;
5383 dst : S4(write);
5384 src : S3(read);
5385 DECODE : S0; // any decoder
5386 ALU : S3; // any alu
5387 %}
5389 // Long ALU reg-reg operation
5390 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
5391 %{
5392 instruction_count(2);
5393 dst : S4(write);
5394 src : S3(read);
5395 DECODE : S0(2); // any 2 decoders
5396 ALU : S3(2); // both alus
5397 %}
5399 // Integer ALU reg-reg operation
5400 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
5401 %{
5402 single_instruction;
5403 dst : S4(write);
5404 src : S3(read);
5405 D0 : S0; // big decoder only
5406 ALU : S3; // any alu
5407 %}
5409 // Long ALU reg-reg operation
5410 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
5411 %{
5412 instruction_count(2);
5413 dst : S4(write);
5414 src : S3(read);
5415 D0 : S0(2); // big decoder only; twice
5416 ALU : S3(2); // both alus
5417 %}
5419 // Integer ALU reg-mem operation
5420 pipe_class ialu_reg_mem(rRegI dst, memory mem)
5421 %{
5422 single_instruction;
5423 dst : S5(write);
5424 mem : S3(read);
5425 D0 : S0; // big decoder only
5426 ALU : S4; // any alu
5427 MEM : S3; // any mem
5428 %}
5430 // Integer mem operation (prefetch)
5431 pipe_class ialu_mem(memory mem)
5432 %{
5433 single_instruction;
5434 mem : S3(read);
5435 D0 : S0; // big decoder only
5436 MEM : S3; // any mem
5437 %}
5439 // Integer Store to Memory
5440 pipe_class ialu_mem_reg(memory mem, rRegI src)
5441 %{
5442 single_instruction;
5443 mem : S3(read);
5444 src : S5(read);
5445 D0 : S0; // big decoder only
5446 ALU : S4; // any alu
5447 MEM : S3;
5448 %}
5450 // // Long Store to Memory
5451 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
5452 // %{
5453 // instruction_count(2);
5454 // mem : S3(read);
5455 // src : S5(read);
5456 // D0 : S0(2); // big decoder only; twice
5457 // ALU : S4(2); // any 2 alus
5458 // MEM : S3(2); // Both mems
5459 // %}
5461 // Integer Store to Memory
5462 pipe_class ialu_mem_imm(memory mem)
5463 %{
5464 single_instruction;
5465 mem : S3(read);
5466 D0 : S0; // big decoder only
5467 ALU : S4; // any alu
5468 MEM : S3;
5469 %}
5471 // Integer ALU0 reg-reg operation
5472 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
5473 %{
5474 single_instruction;
5475 dst : S4(write);
5476 src : S3(read);
5477 D0 : S0; // Big decoder only
5478 ALU0 : S3; // only alu0
5479 %}
5481 // Integer ALU0 reg-mem operation
5482 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
5483 %{
5484 single_instruction;
5485 dst : S5(write);
5486 mem : S3(read);
5487 D0 : S0; // big decoder only
5488 ALU0 : S4; // ALU0 only
5489 MEM : S3; // any mem
5490 %}
5492 // Integer ALU reg-reg operation
5493 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
5494 %{
5495 single_instruction;
5496 cr : S4(write);
5497 src1 : S3(read);
5498 src2 : S3(read);
5499 DECODE : S0; // any decoder
5500 ALU : S3; // any alu
5501 %}
5503 // Integer ALU reg-imm operation
5504 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
5505 %{
5506 single_instruction;
5507 cr : S4(write);
5508 src1 : S3(read);
5509 DECODE : S0; // any decoder
5510 ALU : S3; // any alu
5511 %}
5513 // Integer ALU reg-mem operation
5514 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
5515 %{
5516 single_instruction;
5517 cr : S4(write);
5518 src1 : S3(read);
5519 src2 : S3(read);
5520 D0 : S0; // big decoder only
5521 ALU : S4; // any alu
5522 MEM : S3;
5523 %}
5525 // Conditional move reg-reg
5526 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
5527 %{
5528 instruction_count(4);
5529 y : S4(read);
5530 q : S3(read);
5531 p : S3(read);
5532 DECODE : S0(4); // any decoder
5533 %}
5535 // Conditional move reg-reg
5536 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
5537 %{
5538 single_instruction;
5539 dst : S4(write);
5540 src : S3(read);
5541 cr : S3(read);
5542 DECODE : S0; // any decoder
5543 %}
5545 // Conditional move reg-mem
5546 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
5547 %{
5548 single_instruction;
5549 dst : S4(write);
5550 src : S3(read);
5551 cr : S3(read);
5552 DECODE : S0; // any decoder
5553 MEM : S3;
5554 %}
5556 // Conditional move reg-reg long
5557 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
5558 %{
5559 single_instruction;
5560 dst : S4(write);
5561 src : S3(read);
5562 cr : S3(read);
5563 DECODE : S0(2); // any 2 decoders
5564 %}
5566 // XXX
5567 // // Conditional move double reg-reg
5568 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
5569 // %{
5570 // single_instruction;
5571 // dst : S4(write);
5572 // src : S3(read);
5573 // cr : S3(read);
5574 // DECODE : S0; // any decoder
5575 // %}
5577 // Float reg-reg operation
5578 pipe_class fpu_reg(regD dst)
5579 %{
5580 instruction_count(2);
5581 dst : S3(read);
5582 DECODE : S0(2); // any 2 decoders
5583 FPU : S3;
5584 %}
5586 // Float reg-reg operation
5587 pipe_class fpu_reg_reg(regD dst, regD src)
5588 %{
5589 instruction_count(2);
5590 dst : S4(write);
5591 src : S3(read);
5592 DECODE : S0(2); // any 2 decoders
5593 FPU : S3;
5594 %}
5596 // Float reg-reg operation
5597 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
5598 %{
5599 instruction_count(3);
5600 dst : S4(write);
5601 src1 : S3(read);
5602 src2 : S3(read);
5603 DECODE : S0(3); // any 3 decoders
5604 FPU : S3(2);
5605 %}
5607 // Float reg-reg operation
5608 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
5609 %{
5610 instruction_count(4);
5611 dst : S4(write);
5612 src1 : S3(read);
5613 src2 : S3(read);
5614 src3 : S3(read);
5615 DECODE : S0(4); // any 3 decoders
5616 FPU : S3(2);
5617 %}
5619 // Float reg-reg operation
5620 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
5621 %{
5622 instruction_count(4);
5623 dst : S4(write);
5624 src1 : S3(read);
5625 src2 : S3(read);
5626 src3 : S3(read);
5627 DECODE : S1(3); // any 3 decoders
5628 D0 : S0; // Big decoder only
5629 FPU : S3(2);
5630 MEM : S3;
5631 %}
5633 // Float reg-mem operation
5634 pipe_class fpu_reg_mem(regD dst, memory mem)
5635 %{
5636 instruction_count(2);
5637 dst : S5(write);
5638 mem : S3(read);
5639 D0 : S0; // big decoder only
5640 DECODE : S1; // any decoder for FPU POP
5641 FPU : S4;
5642 MEM : S3; // any mem
5643 %}
5645 // Float reg-mem operation
5646 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
5647 %{
5648 instruction_count(3);
5649 dst : S5(write);
5650 src1 : S3(read);
5651 mem : S3(read);
5652 D0 : S0; // big decoder only
5653 DECODE : S1(2); // any decoder for FPU POP
5654 FPU : S4;
5655 MEM : S3; // any mem
5656 %}
5658 // Float mem-reg operation
5659 pipe_class fpu_mem_reg(memory mem, regD src)
5660 %{
5661 instruction_count(2);
5662 src : S5(read);
5663 mem : S3(read);
5664 DECODE : S0; // any decoder for FPU PUSH
5665 D0 : S1; // big decoder only
5666 FPU : S4;
5667 MEM : S3; // any mem
5668 %}
5670 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
5671 %{
5672 instruction_count(3);
5673 src1 : S3(read);
5674 src2 : S3(read);
5675 mem : S3(read);
5676 DECODE : S0(2); // any decoder for FPU PUSH
5677 D0 : S1; // big decoder only
5678 FPU : S4;
5679 MEM : S3; // any mem
5680 %}
5682 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
5683 %{
5684 instruction_count(3);
5685 src1 : S3(read);
5686 src2 : S3(read);
5687 mem : S4(read);
5688 DECODE : S0; // any decoder for FPU PUSH
5689 D0 : S0(2); // big decoder only
5690 FPU : S4;
5691 MEM : S3(2); // any mem
5692 %}
5694 pipe_class fpu_mem_mem(memory dst, memory src1)
5695 %{
5696 instruction_count(2);
5697 src1 : S3(read);
5698 dst : S4(read);
5699 D0 : S0(2); // big decoder only
5700 MEM : S3(2); // any mem
5701 %}
5703 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
5704 %{
5705 instruction_count(3);
5706 src1 : S3(read);
5707 src2 : S3(read);
5708 dst : S4(read);
5709 D0 : S0(3); // big decoder only
5710 FPU : S4;
5711 MEM : S3(3); // any mem
5712 %}
5714 pipe_class fpu_mem_reg_con(memory mem, regD src1)
5715 %{
5716 instruction_count(3);
5717 src1 : S4(read);
5718 mem : S4(read);
5719 DECODE : S0; // any decoder for FPU PUSH
5720 D0 : S0(2); // big decoder only
5721 FPU : S4;
5722 MEM : S3(2); // any mem
5723 %}
5725 // Float load constant
5726 pipe_class fpu_reg_con(regD dst)
5727 %{
5728 instruction_count(2);
5729 dst : S5(write);
5730 D0 : S0; // big decoder only for the load
5731 DECODE : S1; // any decoder for FPU POP
5732 FPU : S4;
5733 MEM : S3; // any mem
5734 %}
5736 // Float load constant
5737 pipe_class fpu_reg_reg_con(regD dst, regD src)
5738 %{
5739 instruction_count(3);
5740 dst : S5(write);
5741 src : S3(read);
5742 D0 : S0; // big decoder only for the load
5743 DECODE : S1(2); // any decoder for FPU POP
5744 FPU : S4;
5745 MEM : S3; // any mem
5746 %}
5748 // UnConditional branch
5749 pipe_class pipe_jmp(label labl)
5750 %{
5751 single_instruction;
5752 BR : S3;
5753 %}
5755 // Conditional branch
5756 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
5757 %{
5758 single_instruction;
5759 cr : S1(read);
5760 BR : S3;
5761 %}
5763 // Allocation idiom
5764 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
5765 %{
5766 instruction_count(1); force_serialization;
5767 fixed_latency(6);
5768 heap_ptr : S3(read);
5769 DECODE : S0(3);
5770 D0 : S2;
5771 MEM : S3;
5772 ALU : S3(2);
5773 dst : S5(write);
5774 BR : S5;
5775 %}
5777 // Generic big/slow expanded idiom
5778 pipe_class pipe_slow()
5779 %{
5780 instruction_count(10); multiple_bundles; force_serialization;
5781 fixed_latency(100);
5782 D0 : S0(2);
5783 MEM : S3(2);
5784 %}
5786 // The real do-nothing guy
5787 pipe_class empty()
5788 %{
5789 instruction_count(0);
5790 %}
5792 // Define the class for the Nop node
5793 define
5794 %{
5795 MachNop = empty;
5796 %}
5798 %}
5800 //----------INSTRUCTIONS-------------------------------------------------------
5801 //
5802 // match -- States which machine-independent subtree may be replaced
5803 // by this instruction.
5804 // ins_cost -- The estimated cost of this instruction is used by instruction
5805 // selection to identify a minimum cost tree of machine
5806 // instructions that matches a tree of machine-independent
5807 // instructions.
5808 // format -- A string providing the disassembly for this instruction.
5809 // The value of an instruction's operand may be inserted
5810 // by referring to it with a '$' prefix.
5811 // opcode -- Three instruction opcodes may be provided. These are referred
5812 // to within an encode class as $primary, $secondary, and $tertiary
5813 // rrspectively. The primary opcode is commonly used to
5814 // indicate the type of machine instruction, while secondary
5815 // and tertiary are often used for prefix options or addressing
5816 // modes.
5817 // ins_encode -- A list of encode classes with parameters. The encode class
5818 // name must have been defined in an 'enc_class' specification
5819 // in the encode section of the architecture description.
5822 //----------Load/Store/Move Instructions---------------------------------------
5823 //----------Load Instructions--------------------------------------------------
5825 // Load Byte (8 bit signed)
5826 instruct loadB(rRegI dst, memory mem)
5827 %{
5828 match(Set dst (LoadB mem));
5830 ins_cost(125);
5831 format %{ "movsbl $dst, $mem\t# byte" %}
5833 ins_encode %{
5834 __ movsbl($dst$$Register, $mem$$Address);
5835 %}
5837 ins_pipe(ialu_reg_mem);
5838 %}
5840 // Load Byte (8 bit signed) into Long Register
5841 instruct loadB2L(rRegL dst, memory mem)
5842 %{
5843 match(Set dst (ConvI2L (LoadB mem)));
5845 ins_cost(125);
5846 format %{ "movsbq $dst, $mem\t# byte -> long" %}
5848 ins_encode %{
5849 __ movsbq($dst$$Register, $mem$$Address);
5850 %}
5852 ins_pipe(ialu_reg_mem);
5853 %}
5855 // Load Unsigned Byte (8 bit UNsigned)
5856 instruct loadUB(rRegI dst, memory mem)
5857 %{
5858 match(Set dst (LoadUB mem));
5860 ins_cost(125);
5861 format %{ "movzbl $dst, $mem\t# ubyte" %}
5863 ins_encode %{
5864 __ movzbl($dst$$Register, $mem$$Address);
5865 %}
5867 ins_pipe(ialu_reg_mem);
5868 %}
5870 // Load Unsigned Byte (8 bit UNsigned) into Long Register
5871 instruct loadUB2L(rRegL dst, memory mem)
5872 %{
5873 match(Set dst (ConvI2L (LoadUB mem)));
5875 ins_cost(125);
5876 format %{ "movzbq $dst, $mem\t# ubyte -> long" %}
5878 ins_encode %{
5879 __ movzbq($dst$$Register, $mem$$Address);
5880 %}
5882 ins_pipe(ialu_reg_mem);
5883 %}
5885 // Load Unsigned Byte (8 bit UNsigned) with a 8-bit mask into Long Register
5886 instruct loadUB2L_immI8(rRegL dst, memory mem, immI8 mask, rFlagsReg cr) %{
5887 match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
5888 effect(KILL cr);
5890 format %{ "movzbq $dst, $mem\t# ubyte & 8-bit mask -> long\n\t"
5891 "andl $dst, $mask" %}
5892 ins_encode %{
5893 Register Rdst = $dst$$Register;
5894 __ movzbq(Rdst, $mem$$Address);
5895 __ andl(Rdst, $mask$$constant);
5896 %}
5897 ins_pipe(ialu_reg_mem);
5898 %}
5900 // Load Short (16 bit signed)
5901 instruct loadS(rRegI dst, memory mem)
5902 %{
5903 match(Set dst (LoadS mem));
5905 ins_cost(125);
5906 format %{ "movswl $dst, $mem\t# short" %}
5908 ins_encode %{
5909 __ movswl($dst$$Register, $mem$$Address);
5910 %}
5912 ins_pipe(ialu_reg_mem);
5913 %}
5915 // Load Short (16 bit signed) to Byte (8 bit signed)
5916 instruct loadS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
5917 match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
5919 ins_cost(125);
5920 format %{ "movsbl $dst, $mem\t# short -> byte" %}
5921 ins_encode %{
5922 __ movsbl($dst$$Register, $mem$$Address);
5923 %}
5924 ins_pipe(ialu_reg_mem);
5925 %}
5927 // Load Short (16 bit signed) into Long Register
5928 instruct loadS2L(rRegL dst, memory mem)
5929 %{
5930 match(Set dst (ConvI2L (LoadS mem)));
5932 ins_cost(125);
5933 format %{ "movswq $dst, $mem\t# short -> long" %}
5935 ins_encode %{
5936 __ movswq($dst$$Register, $mem$$Address);
5937 %}
5939 ins_pipe(ialu_reg_mem);
5940 %}
5942 // Load Unsigned Short/Char (16 bit UNsigned)
5943 instruct loadUS(rRegI dst, memory mem)
5944 %{
5945 match(Set dst (LoadUS mem));
5947 ins_cost(125);
5948 format %{ "movzwl $dst, $mem\t# ushort/char" %}
5950 ins_encode %{
5951 __ movzwl($dst$$Register, $mem$$Address);
5952 %}
5954 ins_pipe(ialu_reg_mem);
5955 %}
5957 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
5958 instruct loadUS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
5959 match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
5961 ins_cost(125);
5962 format %{ "movsbl $dst, $mem\t# ushort -> byte" %}
5963 ins_encode %{
5964 __ movsbl($dst$$Register, $mem$$Address);
5965 %}
5966 ins_pipe(ialu_reg_mem);
5967 %}
5969 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register
5970 instruct loadUS2L(rRegL dst, memory mem)
5971 %{
5972 match(Set dst (ConvI2L (LoadUS mem)));
5974 ins_cost(125);
5975 format %{ "movzwq $dst, $mem\t# ushort/char -> long" %}
5977 ins_encode %{
5978 __ movzwq($dst$$Register, $mem$$Address);
5979 %}
5981 ins_pipe(ialu_reg_mem);
5982 %}
5984 // Load Unsigned Short/Char (16 bit UNsigned) with mask 0xFF into Long Register
5985 instruct loadUS2L_immI_255(rRegL dst, memory mem, immI_255 mask) %{
5986 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5988 format %{ "movzbq $dst, $mem\t# ushort/char & 0xFF -> long" %}
5989 ins_encode %{
5990 __ movzbq($dst$$Register, $mem$$Address);
5991 %}
5992 ins_pipe(ialu_reg_mem);
5993 %}
5995 // Load Unsigned Short/Char (16 bit UNsigned) with mask into Long Register
5996 instruct loadUS2L_immI16(rRegL dst, memory mem, immI16 mask, rFlagsReg cr) %{
5997 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5998 effect(KILL cr);
6000 format %{ "movzwq $dst, $mem\t# ushort/char & 16-bit mask -> long\n\t"
6001 "andl $dst, $mask" %}
6002 ins_encode %{
6003 Register Rdst = $dst$$Register;
6004 __ movzwq(Rdst, $mem$$Address);
6005 __ andl(Rdst, $mask$$constant);
6006 %}
6007 ins_pipe(ialu_reg_mem);
6008 %}
6010 // Load Integer
6011 instruct loadI(rRegI dst, memory mem)
6012 %{
6013 match(Set dst (LoadI mem));
6015 ins_cost(125);
6016 format %{ "movl $dst, $mem\t# int" %}
6018 ins_encode %{
6019 __ movl($dst$$Register, $mem$$Address);
6020 %}
6022 ins_pipe(ialu_reg_mem);
6023 %}
6025 // Load Integer (32 bit signed) to Byte (8 bit signed)
6026 instruct loadI2B(rRegI dst, memory mem, immI_24 twentyfour) %{
6027 match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
6029 ins_cost(125);
6030 format %{ "movsbl $dst, $mem\t# int -> byte" %}
6031 ins_encode %{
6032 __ movsbl($dst$$Register, $mem$$Address);
6033 %}
6034 ins_pipe(ialu_reg_mem);
6035 %}
6037 // Load Integer (32 bit signed) to Unsigned Byte (8 bit UNsigned)
6038 instruct loadI2UB(rRegI dst, memory mem, immI_255 mask) %{
6039 match(Set dst (AndI (LoadI mem) mask));
6041 ins_cost(125);
6042 format %{ "movzbl $dst, $mem\t# int -> ubyte" %}
6043 ins_encode %{
6044 __ movzbl($dst$$Register, $mem$$Address);
6045 %}
6046 ins_pipe(ialu_reg_mem);
6047 %}
6049 // Load Integer (32 bit signed) to Short (16 bit signed)
6050 instruct loadI2S(rRegI dst, memory mem, immI_16 sixteen) %{
6051 match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
6053 ins_cost(125);
6054 format %{ "movswl $dst, $mem\t# int -> short" %}
6055 ins_encode %{
6056 __ movswl($dst$$Register, $mem$$Address);
6057 %}
6058 ins_pipe(ialu_reg_mem);
6059 %}
6061 // Load Integer (32 bit signed) to Unsigned Short/Char (16 bit UNsigned)
6062 instruct loadI2US(rRegI dst, memory mem, immI_65535 mask) %{
6063 match(Set dst (AndI (LoadI mem) mask));
6065 ins_cost(125);
6066 format %{ "movzwl $dst, $mem\t# int -> ushort/char" %}
6067 ins_encode %{
6068 __ movzwl($dst$$Register, $mem$$Address);
6069 %}
6070 ins_pipe(ialu_reg_mem);
6071 %}
6073 // Load Integer into Long Register
6074 instruct loadI2L(rRegL dst, memory mem)
6075 %{
6076 match(Set dst (ConvI2L (LoadI mem)));
6078 ins_cost(125);
6079 format %{ "movslq $dst, $mem\t# int -> long" %}
6081 ins_encode %{
6082 __ movslq($dst$$Register, $mem$$Address);
6083 %}
6085 ins_pipe(ialu_reg_mem);
6086 %}
6088 // Load Integer with mask 0xFF into Long Register
6089 instruct loadI2L_immI_255(rRegL dst, memory mem, immI_255 mask) %{
6090 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6092 format %{ "movzbq $dst, $mem\t# int & 0xFF -> long" %}
6093 ins_encode %{
6094 __ movzbq($dst$$Register, $mem$$Address);
6095 %}
6096 ins_pipe(ialu_reg_mem);
6097 %}
6099 // Load Integer with mask 0xFFFF into Long Register
6100 instruct loadI2L_immI_65535(rRegL dst, memory mem, immI_65535 mask) %{
6101 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6103 format %{ "movzwq $dst, $mem\t# int & 0xFFFF -> long" %}
6104 ins_encode %{
6105 __ movzwq($dst$$Register, $mem$$Address);
6106 %}
6107 ins_pipe(ialu_reg_mem);
6108 %}
6110 // Load Integer with a 32-bit mask into Long Register
6111 instruct loadI2L_immI(rRegL dst, memory mem, immI mask, rFlagsReg cr) %{
6112 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6113 effect(KILL cr);
6115 format %{ "movl $dst, $mem\t# int & 32-bit mask -> long\n\t"
6116 "andl $dst, $mask" %}
6117 ins_encode %{
6118 Register Rdst = $dst$$Register;
6119 __ movl(Rdst, $mem$$Address);
6120 __ andl(Rdst, $mask$$constant);
6121 %}
6122 ins_pipe(ialu_reg_mem);
6123 %}
6125 // Load Unsigned Integer into Long Register
6126 instruct loadUI2L(rRegL dst, memory mem)
6127 %{
6128 match(Set dst (LoadUI2L mem));
6130 ins_cost(125);
6131 format %{ "movl $dst, $mem\t# uint -> long" %}
6133 ins_encode %{
6134 __ movl($dst$$Register, $mem$$Address);
6135 %}
6137 ins_pipe(ialu_reg_mem);
6138 %}
6140 // Load Long
6141 instruct loadL(rRegL dst, memory mem)
6142 %{
6143 match(Set dst (LoadL mem));
6145 ins_cost(125);
6146 format %{ "movq $dst, $mem\t# long" %}
6148 ins_encode %{
6149 __ movq($dst$$Register, $mem$$Address);
6150 %}
6152 ins_pipe(ialu_reg_mem); // XXX
6153 %}
6155 // Load Range
6156 instruct loadRange(rRegI dst, memory mem)
6157 %{
6158 match(Set dst (LoadRange mem));
6160 ins_cost(125); // XXX
6161 format %{ "movl $dst, $mem\t# range" %}
6162 opcode(0x8B);
6163 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6164 ins_pipe(ialu_reg_mem);
6165 %}
6167 // Load Pointer
6168 instruct loadP(rRegP dst, memory mem)
6169 %{
6170 match(Set dst (LoadP mem));
6172 ins_cost(125); // XXX
6173 format %{ "movq $dst, $mem\t# ptr" %}
6174 opcode(0x8B);
6175 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6176 ins_pipe(ialu_reg_mem); // XXX
6177 %}
6179 // Load Compressed Pointer
6180 instruct loadN(rRegN dst, memory mem)
6181 %{
6182 match(Set dst (LoadN mem));
6184 ins_cost(125); // XXX
6185 format %{ "movl $dst, $mem\t# compressed ptr" %}
6186 ins_encode %{
6187 __ movl($dst$$Register, $mem$$Address);
6188 %}
6189 ins_pipe(ialu_reg_mem); // XXX
6190 %}
6193 // Load Klass Pointer
6194 instruct loadKlass(rRegP dst, memory mem)
6195 %{
6196 match(Set dst (LoadKlass mem));
6198 ins_cost(125); // XXX
6199 format %{ "movq $dst, $mem\t# class" %}
6200 opcode(0x8B);
6201 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6202 ins_pipe(ialu_reg_mem); // XXX
6203 %}
6205 // Load narrow Klass Pointer
6206 instruct loadNKlass(rRegN dst, memory mem)
6207 %{
6208 match(Set dst (LoadNKlass mem));
6210 ins_cost(125); // XXX
6211 format %{ "movl $dst, $mem\t# compressed klass ptr" %}
6212 ins_encode %{
6213 __ movl($dst$$Register, $mem$$Address);
6214 %}
6215 ins_pipe(ialu_reg_mem); // XXX
6216 %}
6218 // Load Float
6219 instruct loadF(regF dst, memory mem)
6220 %{
6221 match(Set dst (LoadF mem));
6223 ins_cost(145); // XXX
6224 format %{ "movss $dst, $mem\t# float" %}
6225 opcode(0xF3, 0x0F, 0x10);
6226 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6227 ins_pipe(pipe_slow); // XXX
6228 %}
6230 // Load Double
6231 instruct loadD_partial(regD dst, memory mem)
6232 %{
6233 predicate(!UseXmmLoadAndClearUpper);
6234 match(Set dst (LoadD mem));
6236 ins_cost(145); // XXX
6237 format %{ "movlpd $dst, $mem\t# double" %}
6238 opcode(0x66, 0x0F, 0x12);
6239 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6240 ins_pipe(pipe_slow); // XXX
6241 %}
6243 instruct loadD(regD dst, memory mem)
6244 %{
6245 predicate(UseXmmLoadAndClearUpper);
6246 match(Set dst (LoadD mem));
6248 ins_cost(145); // XXX
6249 format %{ "movsd $dst, $mem\t# double" %}
6250 opcode(0xF2, 0x0F, 0x10);
6251 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6252 ins_pipe(pipe_slow); // XXX
6253 %}
6255 // Load Aligned Packed Byte to XMM register
6256 instruct loadA8B(regD dst, memory mem) %{
6257 match(Set dst (Load8B mem));
6258 ins_cost(125);
6259 format %{ "MOVQ $dst,$mem\t! packed8B" %}
6260 ins_encode( movq_ld(dst, mem));
6261 ins_pipe( pipe_slow );
6262 %}
6264 // Load Aligned Packed Short to XMM register
6265 instruct loadA4S(regD dst, memory mem) %{
6266 match(Set dst (Load4S mem));
6267 ins_cost(125);
6268 format %{ "MOVQ $dst,$mem\t! packed4S" %}
6269 ins_encode( movq_ld(dst, mem));
6270 ins_pipe( pipe_slow );
6271 %}
6273 // Load Aligned Packed Char to XMM register
6274 instruct loadA4C(regD dst, memory mem) %{
6275 match(Set dst (Load4C mem));
6276 ins_cost(125);
6277 format %{ "MOVQ $dst,$mem\t! packed4C" %}
6278 ins_encode( movq_ld(dst, mem));
6279 ins_pipe( pipe_slow );
6280 %}
6282 // Load Aligned Packed Integer to XMM register
6283 instruct load2IU(regD dst, memory mem) %{
6284 match(Set dst (Load2I mem));
6285 ins_cost(125);
6286 format %{ "MOVQ $dst,$mem\t! packed2I" %}
6287 ins_encode( movq_ld(dst, mem));
6288 ins_pipe( pipe_slow );
6289 %}
6291 // Load Aligned Packed Single to XMM
6292 instruct loadA2F(regD dst, memory mem) %{
6293 match(Set dst (Load2F mem));
6294 ins_cost(145);
6295 format %{ "MOVQ $dst,$mem\t! packed2F" %}
6296 ins_encode( movq_ld(dst, mem));
6297 ins_pipe( pipe_slow );
6298 %}
6300 // Load Effective Address
6301 instruct leaP8(rRegP dst, indOffset8 mem)
6302 %{
6303 match(Set dst mem);
6305 ins_cost(110); // XXX
6306 format %{ "leaq $dst, $mem\t# ptr 8" %}
6307 opcode(0x8D);
6308 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6309 ins_pipe(ialu_reg_reg_fat);
6310 %}
6312 instruct leaP32(rRegP dst, indOffset32 mem)
6313 %{
6314 match(Set dst mem);
6316 ins_cost(110);
6317 format %{ "leaq $dst, $mem\t# ptr 32" %}
6318 opcode(0x8D);
6319 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6320 ins_pipe(ialu_reg_reg_fat);
6321 %}
6323 // instruct leaPIdx(rRegP dst, indIndex mem)
6324 // %{
6325 // match(Set dst mem);
6327 // ins_cost(110);
6328 // format %{ "leaq $dst, $mem\t# ptr idx" %}
6329 // opcode(0x8D);
6330 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6331 // ins_pipe(ialu_reg_reg_fat);
6332 // %}
6334 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
6335 %{
6336 match(Set dst mem);
6338 ins_cost(110);
6339 format %{ "leaq $dst, $mem\t# ptr idxoff" %}
6340 opcode(0x8D);
6341 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6342 ins_pipe(ialu_reg_reg_fat);
6343 %}
6345 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
6346 %{
6347 match(Set dst mem);
6349 ins_cost(110);
6350 format %{ "leaq $dst, $mem\t# ptr idxscale" %}
6351 opcode(0x8D);
6352 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6353 ins_pipe(ialu_reg_reg_fat);
6354 %}
6356 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
6357 %{
6358 match(Set dst mem);
6360 ins_cost(110);
6361 format %{ "leaq $dst, $mem\t# ptr idxscaleoff" %}
6362 opcode(0x8D);
6363 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6364 ins_pipe(ialu_reg_reg_fat);
6365 %}
6367 instruct leaPPosIdxScaleOff(rRegP dst, indPosIndexScaleOffset mem)
6368 %{
6369 match(Set dst mem);
6371 ins_cost(110);
6372 format %{ "leaq $dst, $mem\t# ptr posidxscaleoff" %}
6373 opcode(0x8D);
6374 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6375 ins_pipe(ialu_reg_reg_fat);
6376 %}
6378 // Load Effective Address which uses Narrow (32-bits) oop
6379 instruct leaPCompressedOopOffset(rRegP dst, indCompressedOopOffset mem)
6380 %{
6381 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
6382 match(Set dst mem);
6384 ins_cost(110);
6385 format %{ "leaq $dst, $mem\t# ptr compressedoopoff32" %}
6386 opcode(0x8D);
6387 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6388 ins_pipe(ialu_reg_reg_fat);
6389 %}
6391 instruct leaP8Narrow(rRegP dst, indOffset8Narrow mem)
6392 %{
6393 predicate(Universe::narrow_oop_shift() == 0);
6394 match(Set dst mem);
6396 ins_cost(110); // XXX
6397 format %{ "leaq $dst, $mem\t# ptr off8narrow" %}
6398 opcode(0x8D);
6399 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6400 ins_pipe(ialu_reg_reg_fat);
6401 %}
6403 instruct leaP32Narrow(rRegP dst, indOffset32Narrow mem)
6404 %{
6405 predicate(Universe::narrow_oop_shift() == 0);
6406 match(Set dst mem);
6408 ins_cost(110);
6409 format %{ "leaq $dst, $mem\t# ptr off32narrow" %}
6410 opcode(0x8D);
6411 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6412 ins_pipe(ialu_reg_reg_fat);
6413 %}
6415 instruct leaPIdxOffNarrow(rRegP dst, indIndexOffsetNarrow mem)
6416 %{
6417 predicate(Universe::narrow_oop_shift() == 0);
6418 match(Set dst mem);
6420 ins_cost(110);
6421 format %{ "leaq $dst, $mem\t# ptr idxoffnarrow" %}
6422 opcode(0x8D);
6423 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6424 ins_pipe(ialu_reg_reg_fat);
6425 %}
6427 instruct leaPIdxScaleNarrow(rRegP dst, indIndexScaleNarrow mem)
6428 %{
6429 predicate(Universe::narrow_oop_shift() == 0);
6430 match(Set dst mem);
6432 ins_cost(110);
6433 format %{ "leaq $dst, $mem\t# ptr idxscalenarrow" %}
6434 opcode(0x8D);
6435 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6436 ins_pipe(ialu_reg_reg_fat);
6437 %}
6439 instruct leaPIdxScaleOffNarrow(rRegP dst, indIndexScaleOffsetNarrow mem)
6440 %{
6441 predicate(Universe::narrow_oop_shift() == 0);
6442 match(Set dst mem);
6444 ins_cost(110);
6445 format %{ "leaq $dst, $mem\t# ptr idxscaleoffnarrow" %}
6446 opcode(0x8D);
6447 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6448 ins_pipe(ialu_reg_reg_fat);
6449 %}
6451 instruct leaPPosIdxScaleOffNarrow(rRegP dst, indPosIndexScaleOffsetNarrow mem)
6452 %{
6453 predicate(Universe::narrow_oop_shift() == 0);
6454 match(Set dst mem);
6456 ins_cost(110);
6457 format %{ "leaq $dst, $mem\t# ptr posidxscaleoffnarrow" %}
6458 opcode(0x8D);
6459 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6460 ins_pipe(ialu_reg_reg_fat);
6461 %}
6463 instruct loadConI(rRegI dst, immI src)
6464 %{
6465 match(Set dst src);
6467 format %{ "movl $dst, $src\t# int" %}
6468 ins_encode(load_immI(dst, src));
6469 ins_pipe(ialu_reg_fat); // XXX
6470 %}
6472 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
6473 %{
6474 match(Set dst src);
6475 effect(KILL cr);
6477 ins_cost(50);
6478 format %{ "xorl $dst, $dst\t# int" %}
6479 opcode(0x33); /* + rd */
6480 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6481 ins_pipe(ialu_reg);
6482 %}
6484 instruct loadConL(rRegL dst, immL src)
6485 %{
6486 match(Set dst src);
6488 ins_cost(150);
6489 format %{ "movq $dst, $src\t# long" %}
6490 ins_encode(load_immL(dst, src));
6491 ins_pipe(ialu_reg);
6492 %}
6494 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
6495 %{
6496 match(Set dst src);
6497 effect(KILL cr);
6499 ins_cost(50);
6500 format %{ "xorl $dst, $dst\t# long" %}
6501 opcode(0x33); /* + rd */
6502 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6503 ins_pipe(ialu_reg); // XXX
6504 %}
6506 instruct loadConUL32(rRegL dst, immUL32 src)
6507 %{
6508 match(Set dst src);
6510 ins_cost(60);
6511 format %{ "movl $dst, $src\t# long (unsigned 32-bit)" %}
6512 ins_encode(load_immUL32(dst, src));
6513 ins_pipe(ialu_reg);
6514 %}
6516 instruct loadConL32(rRegL dst, immL32 src)
6517 %{
6518 match(Set dst src);
6520 ins_cost(70);
6521 format %{ "movq $dst, $src\t# long (32-bit)" %}
6522 ins_encode(load_immL32(dst, src));
6523 ins_pipe(ialu_reg);
6524 %}
6526 instruct loadConP(rRegP dst, immP con) %{
6527 match(Set dst con);
6529 format %{ "movq $dst, $con\t# ptr" %}
6530 ins_encode(load_immP(dst, con));
6531 ins_pipe(ialu_reg_fat); // XXX
6532 %}
6534 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
6535 %{
6536 match(Set dst src);
6537 effect(KILL cr);
6539 ins_cost(50);
6540 format %{ "xorl $dst, $dst\t# ptr" %}
6541 opcode(0x33); /* + rd */
6542 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6543 ins_pipe(ialu_reg);
6544 %}
6546 instruct loadConP_poll(rRegP dst, immP_poll src) %{
6547 match(Set dst src);
6548 format %{ "movq $dst, $src\t!ptr" %}
6549 ins_encode %{
6550 AddressLiteral polling_page(os::get_polling_page(), relocInfo::poll_type);
6551 __ lea($dst$$Register, polling_page);
6552 %}
6553 ins_pipe(ialu_reg_fat);
6554 %}
6556 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
6557 %{
6558 match(Set dst src);
6559 effect(KILL cr);
6561 ins_cost(60);
6562 format %{ "movl $dst, $src\t# ptr (positive 32-bit)" %}
6563 ins_encode(load_immP31(dst, src));
6564 ins_pipe(ialu_reg);
6565 %}
6567 instruct loadConF(regF dst, immF con) %{
6568 match(Set dst con);
6569 ins_cost(125);
6570 format %{ "movss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
6571 ins_encode %{
6572 __ movflt($dst$$XMMRegister, $constantaddress($con));
6573 %}
6574 ins_pipe(pipe_slow);
6575 %}
6577 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
6578 match(Set dst src);
6579 effect(KILL cr);
6580 format %{ "xorq $dst, $src\t# compressed NULL ptr" %}
6581 ins_encode %{
6582 __ xorq($dst$$Register, $dst$$Register);
6583 %}
6584 ins_pipe(ialu_reg);
6585 %}
6587 instruct loadConN(rRegN dst, immN src) %{
6588 match(Set dst src);
6590 ins_cost(125);
6591 format %{ "movl $dst, $src\t# compressed ptr" %}
6592 ins_encode %{
6593 address con = (address)$src$$constant;
6594 if (con == NULL) {
6595 ShouldNotReachHere();
6596 } else {
6597 __ set_narrow_oop($dst$$Register, (jobject)$src$$constant);
6598 }
6599 %}
6600 ins_pipe(ialu_reg_fat); // XXX
6601 %}
6603 instruct loadConF0(regF dst, immF0 src)
6604 %{
6605 match(Set dst src);
6606 ins_cost(100);
6608 format %{ "xorps $dst, $dst\t# float 0.0" %}
6609 opcode(0x0F, 0x57);
6610 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
6611 ins_pipe(pipe_slow);
6612 %}
6614 // Use the same format since predicate() can not be used here.
6615 instruct loadConD(regD dst, immD con) %{
6616 match(Set dst con);
6617 ins_cost(125);
6618 format %{ "movsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
6619 ins_encode %{
6620 __ movdbl($dst$$XMMRegister, $constantaddress($con));
6621 %}
6622 ins_pipe(pipe_slow);
6623 %}
6625 instruct loadConD0(regD dst, immD0 src)
6626 %{
6627 match(Set dst src);
6628 ins_cost(100);
6630 format %{ "xorpd $dst, $dst\t# double 0.0" %}
6631 opcode(0x66, 0x0F, 0x57);
6632 ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
6633 ins_pipe(pipe_slow);
6634 %}
6636 instruct loadSSI(rRegI dst, stackSlotI src)
6637 %{
6638 match(Set dst src);
6640 ins_cost(125);
6641 format %{ "movl $dst, $src\t# int stk" %}
6642 opcode(0x8B);
6643 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
6644 ins_pipe(ialu_reg_mem);
6645 %}
6647 instruct loadSSL(rRegL dst, stackSlotL src)
6648 %{
6649 match(Set dst src);
6651 ins_cost(125);
6652 format %{ "movq $dst, $src\t# long stk" %}
6653 opcode(0x8B);
6654 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6655 ins_pipe(ialu_reg_mem);
6656 %}
6658 instruct loadSSP(rRegP dst, stackSlotP src)
6659 %{
6660 match(Set dst src);
6662 ins_cost(125);
6663 format %{ "movq $dst, $src\t# ptr stk" %}
6664 opcode(0x8B);
6665 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6666 ins_pipe(ialu_reg_mem);
6667 %}
6669 instruct loadSSF(regF dst, stackSlotF src)
6670 %{
6671 match(Set dst src);
6673 ins_cost(125);
6674 format %{ "movss $dst, $src\t# float stk" %}
6675 opcode(0xF3, 0x0F, 0x10);
6676 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
6677 ins_pipe(pipe_slow); // XXX
6678 %}
6680 // Use the same format since predicate() can not be used here.
6681 instruct loadSSD(regD dst, stackSlotD src)
6682 %{
6683 match(Set dst src);
6685 ins_cost(125);
6686 format %{ "movsd $dst, $src\t# double stk" %}
6687 ins_encode %{
6688 __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
6689 %}
6690 ins_pipe(pipe_slow); // XXX
6691 %}
6693 // Prefetch instructions.
6694 // Must be safe to execute with invalid address (cannot fault).
6696 instruct prefetchr( memory mem ) %{
6697 predicate(ReadPrefetchInstr==3);
6698 match(PrefetchRead mem);
6699 ins_cost(125);
6701 format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
6702 opcode(0x0F, 0x0D); /* Opcode 0F 0D /0 */
6703 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6704 ins_pipe(ialu_mem);
6705 %}
6707 instruct prefetchrNTA( memory mem ) %{
6708 predicate(ReadPrefetchInstr==0);
6709 match(PrefetchRead mem);
6710 ins_cost(125);
6712 format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
6713 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6714 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6715 ins_pipe(ialu_mem);
6716 %}
6718 instruct prefetchrT0( memory mem ) %{
6719 predicate(ReadPrefetchInstr==1);
6720 match(PrefetchRead mem);
6721 ins_cost(125);
6723 format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
6724 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6725 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6726 ins_pipe(ialu_mem);
6727 %}
6729 instruct prefetchrT2( memory mem ) %{
6730 predicate(ReadPrefetchInstr==2);
6731 match(PrefetchRead mem);
6732 ins_cost(125);
6734 format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
6735 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6736 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6737 ins_pipe(ialu_mem);
6738 %}
6740 instruct prefetchw( memory mem ) %{
6741 predicate(AllocatePrefetchInstr==3);
6742 match(PrefetchWrite mem);
6743 ins_cost(125);
6745 format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
6746 opcode(0x0F, 0x0D); /* Opcode 0F 0D /1 */
6747 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6748 ins_pipe(ialu_mem);
6749 %}
6751 instruct prefetchwNTA( memory mem ) %{
6752 predicate(AllocatePrefetchInstr==0);
6753 match(PrefetchWrite mem);
6754 ins_cost(125);
6756 format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
6757 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6758 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6759 ins_pipe(ialu_mem);
6760 %}
6762 instruct prefetchwT0( memory mem ) %{
6763 predicate(AllocatePrefetchInstr==1);
6764 match(PrefetchWrite mem);
6765 ins_cost(125);
6767 format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
6768 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6769 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6770 ins_pipe(ialu_mem);
6771 %}
6773 instruct prefetchwT2( memory mem ) %{
6774 predicate(AllocatePrefetchInstr==2);
6775 match(PrefetchWrite mem);
6776 ins_cost(125);
6778 format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
6779 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6780 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6781 ins_pipe(ialu_mem);
6782 %}
6784 //----------Store Instructions-------------------------------------------------
6786 // Store Byte
6787 instruct storeB(memory mem, rRegI src)
6788 %{
6789 match(Set mem (StoreB mem src));
6791 ins_cost(125); // XXX
6792 format %{ "movb $mem, $src\t# byte" %}
6793 opcode(0x88);
6794 ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
6795 ins_pipe(ialu_mem_reg);
6796 %}
6798 // Store Char/Short
6799 instruct storeC(memory mem, rRegI src)
6800 %{
6801 match(Set mem (StoreC mem src));
6803 ins_cost(125); // XXX
6804 format %{ "movw $mem, $src\t# char/short" %}
6805 opcode(0x89);
6806 ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6807 ins_pipe(ialu_mem_reg);
6808 %}
6810 // Store Integer
6811 instruct storeI(memory mem, rRegI src)
6812 %{
6813 match(Set mem (StoreI mem src));
6815 ins_cost(125); // XXX
6816 format %{ "movl $mem, $src\t# int" %}
6817 opcode(0x89);
6818 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6819 ins_pipe(ialu_mem_reg);
6820 %}
6822 // Store Long
6823 instruct storeL(memory mem, rRegL src)
6824 %{
6825 match(Set mem (StoreL mem src));
6827 ins_cost(125); // XXX
6828 format %{ "movq $mem, $src\t# long" %}
6829 opcode(0x89);
6830 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6831 ins_pipe(ialu_mem_reg); // XXX
6832 %}
6834 // Store Pointer
6835 instruct storeP(memory mem, any_RegP src)
6836 %{
6837 match(Set mem (StoreP mem src));
6839 ins_cost(125); // XXX
6840 format %{ "movq $mem, $src\t# ptr" %}
6841 opcode(0x89);
6842 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6843 ins_pipe(ialu_mem_reg);
6844 %}
6846 instruct storeImmP0(memory mem, immP0 zero)
6847 %{
6848 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6849 match(Set mem (StoreP mem zero));
6851 ins_cost(125); // XXX
6852 format %{ "movq $mem, R12\t# ptr (R12_heapbase==0)" %}
6853 ins_encode %{
6854 __ movq($mem$$Address, r12);
6855 %}
6856 ins_pipe(ialu_mem_reg);
6857 %}
6859 // Store NULL Pointer, mark word, or other simple pointer constant.
6860 instruct storeImmP(memory mem, immP31 src)
6861 %{
6862 match(Set mem (StoreP mem src));
6864 ins_cost(150); // XXX
6865 format %{ "movq $mem, $src\t# ptr" %}
6866 opcode(0xC7); /* C7 /0 */
6867 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6868 ins_pipe(ialu_mem_imm);
6869 %}
6871 // Store Compressed Pointer
6872 instruct storeN(memory mem, rRegN src)
6873 %{
6874 match(Set mem (StoreN mem src));
6876 ins_cost(125); // XXX
6877 format %{ "movl $mem, $src\t# compressed ptr" %}
6878 ins_encode %{
6879 __ movl($mem$$Address, $src$$Register);
6880 %}
6881 ins_pipe(ialu_mem_reg);
6882 %}
6884 instruct storeImmN0(memory mem, immN0 zero)
6885 %{
6886 predicate(Universe::narrow_oop_base() == NULL);
6887 match(Set mem (StoreN mem zero));
6889 ins_cost(125); // XXX
6890 format %{ "movl $mem, R12\t# compressed ptr (R12_heapbase==0)" %}
6891 ins_encode %{
6892 __ movl($mem$$Address, r12);
6893 %}
6894 ins_pipe(ialu_mem_reg);
6895 %}
6897 instruct storeImmN(memory mem, immN src)
6898 %{
6899 match(Set mem (StoreN mem src));
6901 ins_cost(150); // XXX
6902 format %{ "movl $mem, $src\t# compressed ptr" %}
6903 ins_encode %{
6904 address con = (address)$src$$constant;
6905 if (con == NULL) {
6906 __ movl($mem$$Address, (int32_t)0);
6907 } else {
6908 __ set_narrow_oop($mem$$Address, (jobject)$src$$constant);
6909 }
6910 %}
6911 ins_pipe(ialu_mem_imm);
6912 %}
6914 // Store Integer Immediate
6915 instruct storeImmI0(memory mem, immI0 zero)
6916 %{
6917 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6918 match(Set mem (StoreI mem zero));
6920 ins_cost(125); // XXX
6921 format %{ "movl $mem, R12\t# int (R12_heapbase==0)" %}
6922 ins_encode %{
6923 __ movl($mem$$Address, r12);
6924 %}
6925 ins_pipe(ialu_mem_reg);
6926 %}
6928 instruct storeImmI(memory mem, immI src)
6929 %{
6930 match(Set mem (StoreI mem src));
6932 ins_cost(150);
6933 format %{ "movl $mem, $src\t# int" %}
6934 opcode(0xC7); /* C7 /0 */
6935 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6936 ins_pipe(ialu_mem_imm);
6937 %}
6939 // Store Long Immediate
6940 instruct storeImmL0(memory mem, immL0 zero)
6941 %{
6942 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6943 match(Set mem (StoreL mem zero));
6945 ins_cost(125); // XXX
6946 format %{ "movq $mem, R12\t# long (R12_heapbase==0)" %}
6947 ins_encode %{
6948 __ movq($mem$$Address, r12);
6949 %}
6950 ins_pipe(ialu_mem_reg);
6951 %}
6953 instruct storeImmL(memory mem, immL32 src)
6954 %{
6955 match(Set mem (StoreL mem src));
6957 ins_cost(150);
6958 format %{ "movq $mem, $src\t# long" %}
6959 opcode(0xC7); /* C7 /0 */
6960 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
6961 ins_pipe(ialu_mem_imm);
6962 %}
6964 // Store Short/Char Immediate
6965 instruct storeImmC0(memory mem, immI0 zero)
6966 %{
6967 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6968 match(Set mem (StoreC mem zero));
6970 ins_cost(125); // XXX
6971 format %{ "movw $mem, R12\t# short/char (R12_heapbase==0)" %}
6972 ins_encode %{
6973 __ movw($mem$$Address, r12);
6974 %}
6975 ins_pipe(ialu_mem_reg);
6976 %}
6978 instruct storeImmI16(memory mem, immI16 src)
6979 %{
6980 predicate(UseStoreImmI16);
6981 match(Set mem (StoreC mem src));
6983 ins_cost(150);
6984 format %{ "movw $mem, $src\t# short/char" %}
6985 opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
6986 ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
6987 ins_pipe(ialu_mem_imm);
6988 %}
6990 // Store Byte Immediate
6991 instruct storeImmB0(memory mem, immI0 zero)
6992 %{
6993 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6994 match(Set mem (StoreB mem zero));
6996 ins_cost(125); // XXX
6997 format %{ "movb $mem, R12\t# short/char (R12_heapbase==0)" %}
6998 ins_encode %{
6999 __ movb($mem$$Address, r12);
7000 %}
7001 ins_pipe(ialu_mem_reg);
7002 %}
7004 instruct storeImmB(memory mem, immI8 src)
7005 %{
7006 match(Set mem (StoreB mem src));
7008 ins_cost(150); // XXX
7009 format %{ "movb $mem, $src\t# byte" %}
7010 opcode(0xC6); /* C6 /0 */
7011 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7012 ins_pipe(ialu_mem_imm);
7013 %}
7015 // Store Aligned Packed Byte XMM register to memory
7016 instruct storeA8B(memory mem, regD src) %{
7017 match(Set mem (Store8B mem src));
7018 ins_cost(145);
7019 format %{ "MOVQ $mem,$src\t! packed8B" %}
7020 ins_encode( movq_st(mem, src));
7021 ins_pipe( pipe_slow );
7022 %}
7024 // Store Aligned Packed Char/Short XMM register to memory
7025 instruct storeA4C(memory mem, regD src) %{
7026 match(Set mem (Store4C mem src));
7027 ins_cost(145);
7028 format %{ "MOVQ $mem,$src\t! packed4C" %}
7029 ins_encode( movq_st(mem, src));
7030 ins_pipe( pipe_slow );
7031 %}
7033 // Store Aligned Packed Integer XMM register to memory
7034 instruct storeA2I(memory mem, regD src) %{
7035 match(Set mem (Store2I mem src));
7036 ins_cost(145);
7037 format %{ "MOVQ $mem,$src\t! packed2I" %}
7038 ins_encode( movq_st(mem, src));
7039 ins_pipe( pipe_slow );
7040 %}
7042 // Store CMS card-mark Immediate
7043 instruct storeImmCM0_reg(memory mem, immI0 zero)
7044 %{
7045 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7046 match(Set mem (StoreCM mem zero));
7048 ins_cost(125); // XXX
7049 format %{ "movb $mem, R12\t# CMS card-mark byte 0 (R12_heapbase==0)" %}
7050 ins_encode %{
7051 __ movb($mem$$Address, r12);
7052 %}
7053 ins_pipe(ialu_mem_reg);
7054 %}
7056 instruct storeImmCM0(memory mem, immI0 src)
7057 %{
7058 match(Set mem (StoreCM mem src));
7060 ins_cost(150); // XXX
7061 format %{ "movb $mem, $src\t# CMS card-mark byte 0" %}
7062 opcode(0xC6); /* C6 /0 */
7063 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7064 ins_pipe(ialu_mem_imm);
7065 %}
7067 // Store Aligned Packed Single Float XMM register to memory
7068 instruct storeA2F(memory mem, regD src) %{
7069 match(Set mem (Store2F mem src));
7070 ins_cost(145);
7071 format %{ "MOVQ $mem,$src\t! packed2F" %}
7072 ins_encode( movq_st(mem, src));
7073 ins_pipe( pipe_slow );
7074 %}
7076 // Store Float
7077 instruct storeF(memory mem, regF src)
7078 %{
7079 match(Set mem (StoreF mem src));
7081 ins_cost(95); // XXX
7082 format %{ "movss $mem, $src\t# float" %}
7083 opcode(0xF3, 0x0F, 0x11);
7084 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7085 ins_pipe(pipe_slow); // XXX
7086 %}
7088 // Store immediate Float value (it is faster than store from XMM register)
7089 instruct storeF0(memory mem, immF0 zero)
7090 %{
7091 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7092 match(Set mem (StoreF mem zero));
7094 ins_cost(25); // XXX
7095 format %{ "movl $mem, R12\t# float 0. (R12_heapbase==0)" %}
7096 ins_encode %{
7097 __ movl($mem$$Address, r12);
7098 %}
7099 ins_pipe(ialu_mem_reg);
7100 %}
7102 instruct storeF_imm(memory mem, immF src)
7103 %{
7104 match(Set mem (StoreF mem src));
7106 ins_cost(50);
7107 format %{ "movl $mem, $src\t# float" %}
7108 opcode(0xC7); /* C7 /0 */
7109 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7110 ins_pipe(ialu_mem_imm);
7111 %}
7113 // Store Double
7114 instruct storeD(memory mem, regD src)
7115 %{
7116 match(Set mem (StoreD mem src));
7118 ins_cost(95); // XXX
7119 format %{ "movsd $mem, $src\t# double" %}
7120 opcode(0xF2, 0x0F, 0x11);
7121 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7122 ins_pipe(pipe_slow); // XXX
7123 %}
7125 // Store immediate double 0.0 (it is faster than store from XMM register)
7126 instruct storeD0_imm(memory mem, immD0 src)
7127 %{
7128 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
7129 match(Set mem (StoreD mem src));
7131 ins_cost(50);
7132 format %{ "movq $mem, $src\t# double 0." %}
7133 opcode(0xC7); /* C7 /0 */
7134 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7135 ins_pipe(ialu_mem_imm);
7136 %}
7138 instruct storeD0(memory mem, immD0 zero)
7139 %{
7140 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7141 match(Set mem (StoreD mem zero));
7143 ins_cost(25); // XXX
7144 format %{ "movq $mem, R12\t# double 0. (R12_heapbase==0)" %}
7145 ins_encode %{
7146 __ movq($mem$$Address, r12);
7147 %}
7148 ins_pipe(ialu_mem_reg);
7149 %}
7151 instruct storeSSI(stackSlotI dst, rRegI src)
7152 %{
7153 match(Set dst src);
7155 ins_cost(100);
7156 format %{ "movl $dst, $src\t# int stk" %}
7157 opcode(0x89);
7158 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7159 ins_pipe( ialu_mem_reg );
7160 %}
7162 instruct storeSSL(stackSlotL dst, rRegL src)
7163 %{
7164 match(Set dst src);
7166 ins_cost(100);
7167 format %{ "movq $dst, $src\t# long stk" %}
7168 opcode(0x89);
7169 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7170 ins_pipe(ialu_mem_reg);
7171 %}
7173 instruct storeSSP(stackSlotP dst, rRegP src)
7174 %{
7175 match(Set dst src);
7177 ins_cost(100);
7178 format %{ "movq $dst, $src\t# ptr stk" %}
7179 opcode(0x89);
7180 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7181 ins_pipe(ialu_mem_reg);
7182 %}
7184 instruct storeSSF(stackSlotF dst, regF src)
7185 %{
7186 match(Set dst src);
7188 ins_cost(95); // XXX
7189 format %{ "movss $dst, $src\t# float stk" %}
7190 opcode(0xF3, 0x0F, 0x11);
7191 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7192 ins_pipe(pipe_slow); // XXX
7193 %}
7195 instruct storeSSD(stackSlotD dst, regD src)
7196 %{
7197 match(Set dst src);
7199 ins_cost(95); // XXX
7200 format %{ "movsd $dst, $src\t# double stk" %}
7201 opcode(0xF2, 0x0F, 0x11);
7202 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7203 ins_pipe(pipe_slow); // XXX
7204 %}
7206 //----------BSWAP Instructions-------------------------------------------------
7207 instruct bytes_reverse_int(rRegI dst) %{
7208 match(Set dst (ReverseBytesI dst));
7210 format %{ "bswapl $dst" %}
7211 opcode(0x0F, 0xC8); /*Opcode 0F /C8 */
7212 ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
7213 ins_pipe( ialu_reg );
7214 %}
7216 instruct bytes_reverse_long(rRegL dst) %{
7217 match(Set dst (ReverseBytesL dst));
7219 format %{ "bswapq $dst" %}
7221 opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
7222 ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
7223 ins_pipe( ialu_reg);
7224 %}
7226 instruct bytes_reverse_unsigned_short(rRegI dst) %{
7227 match(Set dst (ReverseBytesUS dst));
7229 format %{ "bswapl $dst\n\t"
7230 "shrl $dst,16\n\t" %}
7231 ins_encode %{
7232 __ bswapl($dst$$Register);
7233 __ shrl($dst$$Register, 16);
7234 %}
7235 ins_pipe( ialu_reg );
7236 %}
7238 instruct bytes_reverse_short(rRegI dst) %{
7239 match(Set dst (ReverseBytesS dst));
7241 format %{ "bswapl $dst\n\t"
7242 "sar $dst,16\n\t" %}
7243 ins_encode %{
7244 __ bswapl($dst$$Register);
7245 __ sarl($dst$$Register, 16);
7246 %}
7247 ins_pipe( ialu_reg );
7248 %}
7250 //---------- Zeros Count Instructions ------------------------------------------
7252 instruct countLeadingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
7253 predicate(UseCountLeadingZerosInstruction);
7254 match(Set dst (CountLeadingZerosI src));
7255 effect(KILL cr);
7257 format %{ "lzcntl $dst, $src\t# count leading zeros (int)" %}
7258 ins_encode %{
7259 __ lzcntl($dst$$Register, $src$$Register);
7260 %}
7261 ins_pipe(ialu_reg);
7262 %}
7264 instruct countLeadingZerosI_bsr(rRegI dst, rRegI src, rFlagsReg cr) %{
7265 predicate(!UseCountLeadingZerosInstruction);
7266 match(Set dst (CountLeadingZerosI src));
7267 effect(KILL cr);
7269 format %{ "bsrl $dst, $src\t# count leading zeros (int)\n\t"
7270 "jnz skip\n\t"
7271 "movl $dst, -1\n"
7272 "skip:\n\t"
7273 "negl $dst\n\t"
7274 "addl $dst, 31" %}
7275 ins_encode %{
7276 Register Rdst = $dst$$Register;
7277 Register Rsrc = $src$$Register;
7278 Label skip;
7279 __ bsrl(Rdst, Rsrc);
7280 __ jccb(Assembler::notZero, skip);
7281 __ movl(Rdst, -1);
7282 __ bind(skip);
7283 __ negl(Rdst);
7284 __ addl(Rdst, BitsPerInt - 1);
7285 %}
7286 ins_pipe(ialu_reg);
7287 %}
7289 instruct countLeadingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
7290 predicate(UseCountLeadingZerosInstruction);
7291 match(Set dst (CountLeadingZerosL src));
7292 effect(KILL cr);
7294 format %{ "lzcntq $dst, $src\t# count leading zeros (long)" %}
7295 ins_encode %{
7296 __ lzcntq($dst$$Register, $src$$Register);
7297 %}
7298 ins_pipe(ialu_reg);
7299 %}
7301 instruct countLeadingZerosL_bsr(rRegI dst, rRegL src, rFlagsReg cr) %{
7302 predicate(!UseCountLeadingZerosInstruction);
7303 match(Set dst (CountLeadingZerosL src));
7304 effect(KILL cr);
7306 format %{ "bsrq $dst, $src\t# count leading zeros (long)\n\t"
7307 "jnz skip\n\t"
7308 "movl $dst, -1\n"
7309 "skip:\n\t"
7310 "negl $dst\n\t"
7311 "addl $dst, 63" %}
7312 ins_encode %{
7313 Register Rdst = $dst$$Register;
7314 Register Rsrc = $src$$Register;
7315 Label skip;
7316 __ bsrq(Rdst, Rsrc);
7317 __ jccb(Assembler::notZero, skip);
7318 __ movl(Rdst, -1);
7319 __ bind(skip);
7320 __ negl(Rdst);
7321 __ addl(Rdst, BitsPerLong - 1);
7322 %}
7323 ins_pipe(ialu_reg);
7324 %}
7326 instruct countTrailingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
7327 match(Set dst (CountTrailingZerosI src));
7328 effect(KILL cr);
7330 format %{ "bsfl $dst, $src\t# count trailing zeros (int)\n\t"
7331 "jnz done\n\t"
7332 "movl $dst, 32\n"
7333 "done:" %}
7334 ins_encode %{
7335 Register Rdst = $dst$$Register;
7336 Label done;
7337 __ bsfl(Rdst, $src$$Register);
7338 __ jccb(Assembler::notZero, done);
7339 __ movl(Rdst, BitsPerInt);
7340 __ bind(done);
7341 %}
7342 ins_pipe(ialu_reg);
7343 %}
7345 instruct countTrailingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
7346 match(Set dst (CountTrailingZerosL src));
7347 effect(KILL cr);
7349 format %{ "bsfq $dst, $src\t# count trailing zeros (long)\n\t"
7350 "jnz done\n\t"
7351 "movl $dst, 64\n"
7352 "done:" %}
7353 ins_encode %{
7354 Register Rdst = $dst$$Register;
7355 Label done;
7356 __ bsfq(Rdst, $src$$Register);
7357 __ jccb(Assembler::notZero, done);
7358 __ movl(Rdst, BitsPerLong);
7359 __ bind(done);
7360 %}
7361 ins_pipe(ialu_reg);
7362 %}
7365 //---------- Population Count Instructions -------------------------------------
7367 instruct popCountI(rRegI dst, rRegI src) %{
7368 predicate(UsePopCountInstruction);
7369 match(Set dst (PopCountI src));
7371 format %{ "popcnt $dst, $src" %}
7372 ins_encode %{
7373 __ popcntl($dst$$Register, $src$$Register);
7374 %}
7375 ins_pipe(ialu_reg);
7376 %}
7378 instruct popCountI_mem(rRegI dst, memory mem) %{
7379 predicate(UsePopCountInstruction);
7380 match(Set dst (PopCountI (LoadI mem)));
7382 format %{ "popcnt $dst, $mem" %}
7383 ins_encode %{
7384 __ popcntl($dst$$Register, $mem$$Address);
7385 %}
7386 ins_pipe(ialu_reg);
7387 %}
7389 // Note: Long.bitCount(long) returns an int.
7390 instruct popCountL(rRegI dst, rRegL src) %{
7391 predicate(UsePopCountInstruction);
7392 match(Set dst (PopCountL src));
7394 format %{ "popcnt $dst, $src" %}
7395 ins_encode %{
7396 __ popcntq($dst$$Register, $src$$Register);
7397 %}
7398 ins_pipe(ialu_reg);
7399 %}
7401 // Note: Long.bitCount(long) returns an int.
7402 instruct popCountL_mem(rRegI dst, memory mem) %{
7403 predicate(UsePopCountInstruction);
7404 match(Set dst (PopCountL (LoadL mem)));
7406 format %{ "popcnt $dst, $mem" %}
7407 ins_encode %{
7408 __ popcntq($dst$$Register, $mem$$Address);
7409 %}
7410 ins_pipe(ialu_reg);
7411 %}
7414 //----------MemBar Instructions-----------------------------------------------
7415 // Memory barrier flavors
7417 instruct membar_acquire()
7418 %{
7419 match(MemBarAcquire);
7420 ins_cost(0);
7422 size(0);
7423 format %{ "MEMBAR-acquire ! (empty encoding)" %}
7424 ins_encode();
7425 ins_pipe(empty);
7426 %}
7428 instruct membar_acquire_lock()
7429 %{
7430 match(MemBarAcquire);
7431 predicate(Matcher::prior_fast_lock(n));
7432 ins_cost(0);
7434 size(0);
7435 format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7436 ins_encode();
7437 ins_pipe(empty);
7438 %}
7440 instruct membar_release()
7441 %{
7442 match(MemBarRelease);
7443 ins_cost(0);
7445 size(0);
7446 format %{ "MEMBAR-release ! (empty encoding)" %}
7447 ins_encode();
7448 ins_pipe(empty);
7449 %}
7451 instruct membar_release_lock()
7452 %{
7453 match(MemBarRelease);
7454 predicate(Matcher::post_fast_unlock(n));
7455 ins_cost(0);
7457 size(0);
7458 format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7459 ins_encode();
7460 ins_pipe(empty);
7461 %}
7463 instruct membar_volatile(rFlagsReg cr) %{
7464 match(MemBarVolatile);
7465 effect(KILL cr);
7466 ins_cost(400);
7468 format %{
7469 $$template
7470 if (os::is_MP()) {
7471 $$emit$$"lock addl [rsp + #0], 0\t! membar_volatile"
7472 } else {
7473 $$emit$$"MEMBAR-volatile ! (empty encoding)"
7474 }
7475 %}
7476 ins_encode %{
7477 __ membar(Assembler::StoreLoad);
7478 %}
7479 ins_pipe(pipe_slow);
7480 %}
7482 instruct unnecessary_membar_volatile()
7483 %{
7484 match(MemBarVolatile);
7485 predicate(Matcher::post_store_load_barrier(n));
7486 ins_cost(0);
7488 size(0);
7489 format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7490 ins_encode();
7491 ins_pipe(empty);
7492 %}
7494 //----------Move Instructions--------------------------------------------------
7496 instruct castX2P(rRegP dst, rRegL src)
7497 %{
7498 match(Set dst (CastX2P src));
7500 format %{ "movq $dst, $src\t# long->ptr" %}
7501 ins_encode(enc_copy_wide(dst, src));
7502 ins_pipe(ialu_reg_reg); // XXX
7503 %}
7505 instruct castP2X(rRegL dst, rRegP src)
7506 %{
7507 match(Set dst (CastP2X src));
7509 format %{ "movq $dst, $src\t# ptr -> long" %}
7510 ins_encode(enc_copy_wide(dst, src));
7511 ins_pipe(ialu_reg_reg); // XXX
7512 %}
7515 // Convert oop pointer into compressed form
7516 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
7517 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
7518 match(Set dst (EncodeP src));
7519 effect(KILL cr);
7520 format %{ "encode_heap_oop $dst,$src" %}
7521 ins_encode %{
7522 Register s = $src$$Register;
7523 Register d = $dst$$Register;
7524 if (s != d) {
7525 __ movq(d, s);
7526 }
7527 __ encode_heap_oop(d);
7528 %}
7529 ins_pipe(ialu_reg_long);
7530 %}
7532 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
7533 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
7534 match(Set dst (EncodeP src));
7535 effect(KILL cr);
7536 format %{ "encode_heap_oop_not_null $dst,$src" %}
7537 ins_encode %{
7538 __ encode_heap_oop_not_null($dst$$Register, $src$$Register);
7539 %}
7540 ins_pipe(ialu_reg_long);
7541 %}
7543 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
7544 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
7545 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
7546 match(Set dst (DecodeN src));
7547 effect(KILL cr);
7548 format %{ "decode_heap_oop $dst,$src" %}
7549 ins_encode %{
7550 Register s = $src$$Register;
7551 Register d = $dst$$Register;
7552 if (s != d) {
7553 __ movq(d, s);
7554 }
7555 __ decode_heap_oop(d);
7556 %}
7557 ins_pipe(ialu_reg_long);
7558 %}
7560 instruct decodeHeapOop_not_null(rRegP dst, rRegN src, rFlagsReg cr) %{
7561 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
7562 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
7563 match(Set dst (DecodeN src));
7564 effect(KILL cr);
7565 format %{ "decode_heap_oop_not_null $dst,$src" %}
7566 ins_encode %{
7567 Register s = $src$$Register;
7568 Register d = $dst$$Register;
7569 if (s != d) {
7570 __ decode_heap_oop_not_null(d, s);
7571 } else {
7572 __ decode_heap_oop_not_null(d);
7573 }
7574 %}
7575 ins_pipe(ialu_reg_long);
7576 %}
7579 //----------Conditional Move---------------------------------------------------
7580 // Jump
7581 // dummy instruction for generating temp registers
7582 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
7583 match(Jump (LShiftL switch_val shift));
7584 ins_cost(350);
7585 predicate(false);
7586 effect(TEMP dest);
7588 format %{ "leaq $dest, [$constantaddress]\n\t"
7589 "jmp [$dest + $switch_val << $shift]\n\t" %}
7590 ins_encode %{
7591 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
7592 // to do that and the compiler is using that register as one it can allocate.
7593 // So we build it all by hand.
7594 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
7595 // ArrayAddress dispatch(table, index);
7596 Address dispatch($dest$$Register, $switch_val$$Register, (Address::ScaleFactor) $shift$$constant);
7597 __ lea($dest$$Register, $constantaddress);
7598 __ jmp(dispatch);
7599 %}
7600 ins_pipe(pipe_jmp);
7601 ins_pc_relative(1);
7602 %}
7604 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
7605 match(Jump (AddL (LShiftL switch_val shift) offset));
7606 ins_cost(350);
7607 effect(TEMP dest);
7609 format %{ "leaq $dest, [$constantaddress]\n\t"
7610 "jmp [$dest + $switch_val << $shift + $offset]\n\t" %}
7611 ins_encode %{
7612 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
7613 // to do that and the compiler is using that register as one it can allocate.
7614 // So we build it all by hand.
7615 // Address index(noreg, switch_reg, (Address::ScaleFactor) $shift$$constant, (int) $offset$$constant);
7616 // ArrayAddress dispatch(table, index);
7617 Address dispatch($dest$$Register, $switch_val$$Register, (Address::ScaleFactor) $shift$$constant, (int) $offset$$constant);
7618 __ lea($dest$$Register, $constantaddress);
7619 __ jmp(dispatch);
7620 %}
7621 ins_pipe(pipe_jmp);
7622 ins_pc_relative(1);
7623 %}
7625 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
7626 match(Jump switch_val);
7627 ins_cost(350);
7628 effect(TEMP dest);
7630 format %{ "leaq $dest, [$constantaddress]\n\t"
7631 "jmp [$dest + $switch_val]\n\t" %}
7632 ins_encode %{
7633 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
7634 // to do that and the compiler is using that register as one it can allocate.
7635 // So we build it all by hand.
7636 // Address index(noreg, switch_reg, Address::times_1);
7637 // ArrayAddress dispatch(table, index);
7638 Address dispatch($dest$$Register, $switch_val$$Register, Address::times_1);
7639 __ lea($dest$$Register, $constantaddress);
7640 __ jmp(dispatch);
7641 %}
7642 ins_pipe(pipe_jmp);
7643 ins_pc_relative(1);
7644 %}
7646 // Conditional move
7647 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
7648 %{
7649 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7651 ins_cost(200); // XXX
7652 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7653 opcode(0x0F, 0x40);
7654 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7655 ins_pipe(pipe_cmov_reg);
7656 %}
7658 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{
7659 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7661 ins_cost(200); // XXX
7662 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7663 opcode(0x0F, 0x40);
7664 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7665 ins_pipe(pipe_cmov_reg);
7666 %}
7668 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
7669 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7670 ins_cost(200);
7671 expand %{
7672 cmovI_regU(cop, cr, dst, src);
7673 %}
7674 %}
7676 // Conditional move
7677 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
7678 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7680 ins_cost(250); // XXX
7681 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7682 opcode(0x0F, 0x40);
7683 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7684 ins_pipe(pipe_cmov_mem);
7685 %}
7687 // Conditional move
7688 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
7689 %{
7690 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7692 ins_cost(250); // XXX
7693 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7694 opcode(0x0F, 0x40);
7695 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7696 ins_pipe(pipe_cmov_mem);
7697 %}
7699 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
7700 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7701 ins_cost(250);
7702 expand %{
7703 cmovI_memU(cop, cr, dst, src);
7704 %}
7705 %}
7707 // Conditional move
7708 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop)
7709 %{
7710 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7712 ins_cost(200); // XXX
7713 format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %}
7714 opcode(0x0F, 0x40);
7715 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7716 ins_pipe(pipe_cmov_reg);
7717 %}
7719 // Conditional move
7720 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
7721 %{
7722 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7724 ins_cost(200); // XXX
7725 format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %}
7726 opcode(0x0F, 0x40);
7727 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7728 ins_pipe(pipe_cmov_reg);
7729 %}
7731 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
7732 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7733 ins_cost(200);
7734 expand %{
7735 cmovN_regU(cop, cr, dst, src);
7736 %}
7737 %}
7739 // Conditional move
7740 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
7741 %{
7742 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7744 ins_cost(200); // XXX
7745 format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
7746 opcode(0x0F, 0x40);
7747 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7748 ins_pipe(pipe_cmov_reg); // XXX
7749 %}
7751 // Conditional move
7752 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src)
7753 %{
7754 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7756 ins_cost(200); // XXX
7757 format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
7758 opcode(0x0F, 0x40);
7759 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7760 ins_pipe(pipe_cmov_reg); // XXX
7761 %}
7763 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
7764 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7765 ins_cost(200);
7766 expand %{
7767 cmovP_regU(cop, cr, dst, src);
7768 %}
7769 %}
7771 // DISABLED: Requires the ADLC to emit a bottom_type call that
7772 // correctly meets the two pointer arguments; one is an incoming
7773 // register but the other is a memory operand. ALSO appears to
7774 // be buggy with implicit null checks.
7775 //
7776 //// Conditional move
7777 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
7778 //%{
7779 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7780 // ins_cost(250);
7781 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7782 // opcode(0x0F,0x40);
7783 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7784 // ins_pipe( pipe_cmov_mem );
7785 //%}
7786 //
7787 //// Conditional move
7788 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
7789 //%{
7790 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7791 // ins_cost(250);
7792 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7793 // opcode(0x0F,0x40);
7794 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7795 // ins_pipe( pipe_cmov_mem );
7796 //%}
7798 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
7799 %{
7800 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7802 ins_cost(200); // XXX
7803 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7804 opcode(0x0F, 0x40);
7805 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7806 ins_pipe(pipe_cmov_reg); // XXX
7807 %}
7809 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
7810 %{
7811 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7813 ins_cost(200); // XXX
7814 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7815 opcode(0x0F, 0x40);
7816 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7817 ins_pipe(pipe_cmov_mem); // XXX
7818 %}
7820 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
7821 %{
7822 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7824 ins_cost(200); // XXX
7825 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7826 opcode(0x0F, 0x40);
7827 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7828 ins_pipe(pipe_cmov_reg); // XXX
7829 %}
7831 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
7832 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7833 ins_cost(200);
7834 expand %{
7835 cmovL_regU(cop, cr, dst, src);
7836 %}
7837 %}
7839 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
7840 %{
7841 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7843 ins_cost(200); // XXX
7844 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7845 opcode(0x0F, 0x40);
7846 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7847 ins_pipe(pipe_cmov_mem); // XXX
7848 %}
7850 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
7851 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7852 ins_cost(200);
7853 expand %{
7854 cmovL_memU(cop, cr, dst, src);
7855 %}
7856 %}
7858 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
7859 %{
7860 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7862 ins_cost(200); // XXX
7863 format %{ "jn$cop skip\t# signed cmove float\n\t"
7864 "movss $dst, $src\n"
7865 "skip:" %}
7866 ins_encode(enc_cmovf_branch(cop, dst, src));
7867 ins_pipe(pipe_slow);
7868 %}
7870 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
7871 // %{
7872 // match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
7874 // ins_cost(200); // XXX
7875 // format %{ "jn$cop skip\t# signed cmove float\n\t"
7876 // "movss $dst, $src\n"
7877 // "skip:" %}
7878 // ins_encode(enc_cmovf_mem_branch(cop, dst, src));
7879 // ins_pipe(pipe_slow);
7880 // %}
7882 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
7883 %{
7884 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7886 ins_cost(200); // XXX
7887 format %{ "jn$cop skip\t# unsigned cmove float\n\t"
7888 "movss $dst, $src\n"
7889 "skip:" %}
7890 ins_encode(enc_cmovf_branch(cop, dst, src));
7891 ins_pipe(pipe_slow);
7892 %}
7894 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
7895 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
7896 ins_cost(200);
7897 expand %{
7898 cmovF_regU(cop, cr, dst, src);
7899 %}
7900 %}
7902 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
7903 %{
7904 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7906 ins_cost(200); // XXX
7907 format %{ "jn$cop skip\t# signed cmove double\n\t"
7908 "movsd $dst, $src\n"
7909 "skip:" %}
7910 ins_encode(enc_cmovd_branch(cop, dst, src));
7911 ins_pipe(pipe_slow);
7912 %}
7914 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
7915 %{
7916 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7918 ins_cost(200); // XXX
7919 format %{ "jn$cop skip\t# unsigned cmove double\n\t"
7920 "movsd $dst, $src\n"
7921 "skip:" %}
7922 ins_encode(enc_cmovd_branch(cop, dst, src));
7923 ins_pipe(pipe_slow);
7924 %}
7926 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
7927 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
7928 ins_cost(200);
7929 expand %{
7930 cmovD_regU(cop, cr, dst, src);
7931 %}
7932 %}
7934 //----------Arithmetic Instructions--------------------------------------------
7935 //----------Addition Instructions----------------------------------------------
7937 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
7938 %{
7939 match(Set dst (AddI dst src));
7940 effect(KILL cr);
7942 format %{ "addl $dst, $src\t# int" %}
7943 opcode(0x03);
7944 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
7945 ins_pipe(ialu_reg_reg);
7946 %}
7948 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
7949 %{
7950 match(Set dst (AddI dst src));
7951 effect(KILL cr);
7953 format %{ "addl $dst, $src\t# int" %}
7954 opcode(0x81, 0x00); /* /0 id */
7955 ins_encode(OpcSErm(dst, src), Con8or32(src));
7956 ins_pipe( ialu_reg );
7957 %}
7959 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
7960 %{
7961 match(Set dst (AddI dst (LoadI src)));
7962 effect(KILL cr);
7964 ins_cost(125); // XXX
7965 format %{ "addl $dst, $src\t# int" %}
7966 opcode(0x03);
7967 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
7968 ins_pipe(ialu_reg_mem);
7969 %}
7971 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
7972 %{
7973 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7974 effect(KILL cr);
7976 ins_cost(150); // XXX
7977 format %{ "addl $dst, $src\t# int" %}
7978 opcode(0x01); /* Opcode 01 /r */
7979 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7980 ins_pipe(ialu_mem_reg);
7981 %}
7983 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
7984 %{
7985 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
7986 effect(KILL cr);
7988 ins_cost(125); // XXX
7989 format %{ "addl $dst, $src\t# int" %}
7990 opcode(0x81); /* Opcode 81 /0 id */
7991 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
7992 ins_pipe(ialu_mem_imm);
7993 %}
7995 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
7996 %{
7997 predicate(UseIncDec);
7998 match(Set dst (AddI dst src));
7999 effect(KILL cr);
8001 format %{ "incl $dst\t# int" %}
8002 opcode(0xFF, 0x00); // FF /0
8003 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8004 ins_pipe(ialu_reg);
8005 %}
8007 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
8008 %{
8009 predicate(UseIncDec);
8010 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8011 effect(KILL cr);
8013 ins_cost(125); // XXX
8014 format %{ "incl $dst\t# int" %}
8015 opcode(0xFF); /* Opcode FF /0 */
8016 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
8017 ins_pipe(ialu_mem_imm);
8018 %}
8020 // XXX why does that use AddI
8021 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
8022 %{
8023 predicate(UseIncDec);
8024 match(Set dst (AddI dst src));
8025 effect(KILL cr);
8027 format %{ "decl $dst\t# int" %}
8028 opcode(0xFF, 0x01); // FF /1
8029 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8030 ins_pipe(ialu_reg);
8031 %}
8033 // XXX why does that use AddI
8034 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
8035 %{
8036 predicate(UseIncDec);
8037 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8038 effect(KILL cr);
8040 ins_cost(125); // XXX
8041 format %{ "decl $dst\t# int" %}
8042 opcode(0xFF); /* Opcode FF /1 */
8043 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
8044 ins_pipe(ialu_mem_imm);
8045 %}
8047 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
8048 %{
8049 match(Set dst (AddI src0 src1));
8051 ins_cost(110);
8052 format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
8053 opcode(0x8D); /* 0x8D /r */
8054 ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8055 ins_pipe(ialu_reg_reg);
8056 %}
8058 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8059 %{
8060 match(Set dst (AddL dst src));
8061 effect(KILL cr);
8063 format %{ "addq $dst, $src\t# long" %}
8064 opcode(0x03);
8065 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8066 ins_pipe(ialu_reg_reg);
8067 %}
8069 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
8070 %{
8071 match(Set dst (AddL dst src));
8072 effect(KILL cr);
8074 format %{ "addq $dst, $src\t# long" %}
8075 opcode(0x81, 0x00); /* /0 id */
8076 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8077 ins_pipe( ialu_reg );
8078 %}
8080 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8081 %{
8082 match(Set dst (AddL dst (LoadL src)));
8083 effect(KILL cr);
8085 ins_cost(125); // XXX
8086 format %{ "addq $dst, $src\t# long" %}
8087 opcode(0x03);
8088 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8089 ins_pipe(ialu_reg_mem);
8090 %}
8092 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8093 %{
8094 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8095 effect(KILL cr);
8097 ins_cost(150); // XXX
8098 format %{ "addq $dst, $src\t# long" %}
8099 opcode(0x01); /* Opcode 01 /r */
8100 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8101 ins_pipe(ialu_mem_reg);
8102 %}
8104 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8105 %{
8106 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8107 effect(KILL cr);
8109 ins_cost(125); // XXX
8110 format %{ "addq $dst, $src\t# long" %}
8111 opcode(0x81); /* Opcode 81 /0 id */
8112 ins_encode(REX_mem_wide(dst),
8113 OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
8114 ins_pipe(ialu_mem_imm);
8115 %}
8117 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
8118 %{
8119 predicate(UseIncDec);
8120 match(Set dst (AddL dst src));
8121 effect(KILL cr);
8123 format %{ "incq $dst\t# long" %}
8124 opcode(0xFF, 0x00); // FF /0
8125 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8126 ins_pipe(ialu_reg);
8127 %}
8129 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
8130 %{
8131 predicate(UseIncDec);
8132 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8133 effect(KILL cr);
8135 ins_cost(125); // XXX
8136 format %{ "incq $dst\t# long" %}
8137 opcode(0xFF); /* Opcode FF /0 */
8138 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
8139 ins_pipe(ialu_mem_imm);
8140 %}
8142 // XXX why does that use AddL
8143 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
8144 %{
8145 predicate(UseIncDec);
8146 match(Set dst (AddL dst src));
8147 effect(KILL cr);
8149 format %{ "decq $dst\t# long" %}
8150 opcode(0xFF, 0x01); // FF /1
8151 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8152 ins_pipe(ialu_reg);
8153 %}
8155 // XXX why does that use AddL
8156 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
8157 %{
8158 predicate(UseIncDec);
8159 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8160 effect(KILL cr);
8162 ins_cost(125); // XXX
8163 format %{ "decq $dst\t# long" %}
8164 opcode(0xFF); /* Opcode FF /1 */
8165 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
8166 ins_pipe(ialu_mem_imm);
8167 %}
8169 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
8170 %{
8171 match(Set dst (AddL src0 src1));
8173 ins_cost(110);
8174 format %{ "leaq $dst, [$src0 + $src1]\t# long" %}
8175 opcode(0x8D); /* 0x8D /r */
8176 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8177 ins_pipe(ialu_reg_reg);
8178 %}
8180 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
8181 %{
8182 match(Set dst (AddP dst src));
8183 effect(KILL cr);
8185 format %{ "addq $dst, $src\t# ptr" %}
8186 opcode(0x03);
8187 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8188 ins_pipe(ialu_reg_reg);
8189 %}
8191 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
8192 %{
8193 match(Set dst (AddP dst src));
8194 effect(KILL cr);
8196 format %{ "addq $dst, $src\t# ptr" %}
8197 opcode(0x81, 0x00); /* /0 id */
8198 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8199 ins_pipe( ialu_reg );
8200 %}
8202 // XXX addP mem ops ????
8204 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
8205 %{
8206 match(Set dst (AddP src0 src1));
8208 ins_cost(110);
8209 format %{ "leaq $dst, [$src0 + $src1]\t# ptr" %}
8210 opcode(0x8D); /* 0x8D /r */
8211 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
8212 ins_pipe(ialu_reg_reg);
8213 %}
8215 instruct checkCastPP(rRegP dst)
8216 %{
8217 match(Set dst (CheckCastPP dst));
8219 size(0);
8220 format %{ "# checkcastPP of $dst" %}
8221 ins_encode(/* empty encoding */);
8222 ins_pipe(empty);
8223 %}
8225 instruct castPP(rRegP dst)
8226 %{
8227 match(Set dst (CastPP dst));
8229 size(0);
8230 format %{ "# castPP of $dst" %}
8231 ins_encode(/* empty encoding */);
8232 ins_pipe(empty);
8233 %}
8235 instruct castII(rRegI dst)
8236 %{
8237 match(Set dst (CastII dst));
8239 size(0);
8240 format %{ "# castII of $dst" %}
8241 ins_encode(/* empty encoding */);
8242 ins_cost(0);
8243 ins_pipe(empty);
8244 %}
8246 // LoadP-locked same as a regular LoadP when used with compare-swap
8247 instruct loadPLocked(rRegP dst, memory mem)
8248 %{
8249 match(Set dst (LoadPLocked mem));
8251 ins_cost(125); // XXX
8252 format %{ "movq $dst, $mem\t# ptr locked" %}
8253 opcode(0x8B);
8254 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8255 ins_pipe(ialu_reg_mem); // XXX
8256 %}
8258 // LoadL-locked - same as a regular LoadL when used with compare-swap
8259 instruct loadLLocked(rRegL dst, memory mem)
8260 %{
8261 match(Set dst (LoadLLocked mem));
8263 ins_cost(125); // XXX
8264 format %{ "movq $dst, $mem\t# long locked" %}
8265 opcode(0x8B);
8266 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8267 ins_pipe(ialu_reg_mem); // XXX
8268 %}
8270 // Conditional-store of the updated heap-top.
8271 // Used during allocation of the shared heap.
8272 // Sets flags (EQ) on success. Implemented with a CMPXCHG on Intel.
8274 instruct storePConditional(memory heap_top_ptr,
8275 rax_RegP oldval, rRegP newval,
8276 rFlagsReg cr)
8277 %{
8278 match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
8280 format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
8281 "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
8282 opcode(0x0F, 0xB1);
8283 ins_encode(lock_prefix,
8284 REX_reg_mem_wide(newval, heap_top_ptr),
8285 OpcP, OpcS,
8286 reg_mem(newval, heap_top_ptr));
8287 ins_pipe(pipe_cmpxchg);
8288 %}
8290 // Conditional-store of an int value.
8291 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8292 instruct storeIConditional(memory mem, rax_RegI oldval, rRegI newval, rFlagsReg cr)
8293 %{
8294 match(Set cr (StoreIConditional mem (Binary oldval newval)));
8295 effect(KILL oldval);
8297 format %{ "cmpxchgl $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8298 opcode(0x0F, 0xB1);
8299 ins_encode(lock_prefix,
8300 REX_reg_mem(newval, mem),
8301 OpcP, OpcS,
8302 reg_mem(newval, mem));
8303 ins_pipe(pipe_cmpxchg);
8304 %}
8306 // Conditional-store of a long value.
8307 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8308 instruct storeLConditional(memory mem, rax_RegL oldval, rRegL newval, rFlagsReg cr)
8309 %{
8310 match(Set cr (StoreLConditional mem (Binary oldval newval)));
8311 effect(KILL oldval);
8313 format %{ "cmpxchgq $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8314 opcode(0x0F, 0xB1);
8315 ins_encode(lock_prefix,
8316 REX_reg_mem_wide(newval, mem),
8317 OpcP, OpcS,
8318 reg_mem(newval, mem));
8319 ins_pipe(pipe_cmpxchg);
8320 %}
8323 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
8324 instruct compareAndSwapP(rRegI res,
8325 memory mem_ptr,
8326 rax_RegP oldval, rRegP newval,
8327 rFlagsReg cr)
8328 %{
8329 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
8330 effect(KILL cr, KILL oldval);
8332 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8333 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8334 "sete $res\n\t"
8335 "movzbl $res, $res" %}
8336 opcode(0x0F, 0xB1);
8337 ins_encode(lock_prefix,
8338 REX_reg_mem_wide(newval, mem_ptr),
8339 OpcP, OpcS,
8340 reg_mem(newval, mem_ptr),
8341 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8342 REX_reg_breg(res, res), // movzbl
8343 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8344 ins_pipe( pipe_cmpxchg );
8345 %}
8347 instruct compareAndSwapL(rRegI res,
8348 memory mem_ptr,
8349 rax_RegL oldval, rRegL newval,
8350 rFlagsReg cr)
8351 %{
8352 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
8353 effect(KILL cr, KILL oldval);
8355 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8356 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8357 "sete $res\n\t"
8358 "movzbl $res, $res" %}
8359 opcode(0x0F, 0xB1);
8360 ins_encode(lock_prefix,
8361 REX_reg_mem_wide(newval, mem_ptr),
8362 OpcP, OpcS,
8363 reg_mem(newval, mem_ptr),
8364 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8365 REX_reg_breg(res, res), // movzbl
8366 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8367 ins_pipe( pipe_cmpxchg );
8368 %}
8370 instruct compareAndSwapI(rRegI res,
8371 memory mem_ptr,
8372 rax_RegI oldval, rRegI newval,
8373 rFlagsReg cr)
8374 %{
8375 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
8376 effect(KILL cr, KILL oldval);
8378 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8379 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8380 "sete $res\n\t"
8381 "movzbl $res, $res" %}
8382 opcode(0x0F, 0xB1);
8383 ins_encode(lock_prefix,
8384 REX_reg_mem(newval, mem_ptr),
8385 OpcP, OpcS,
8386 reg_mem(newval, mem_ptr),
8387 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8388 REX_reg_breg(res, res), // movzbl
8389 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8390 ins_pipe( pipe_cmpxchg );
8391 %}
8394 instruct compareAndSwapN(rRegI res,
8395 memory mem_ptr,
8396 rax_RegN oldval, rRegN newval,
8397 rFlagsReg cr) %{
8398 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
8399 effect(KILL cr, KILL oldval);
8401 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8402 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8403 "sete $res\n\t"
8404 "movzbl $res, $res" %}
8405 opcode(0x0F, 0xB1);
8406 ins_encode(lock_prefix,
8407 REX_reg_mem(newval, mem_ptr),
8408 OpcP, OpcS,
8409 reg_mem(newval, mem_ptr),
8410 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8411 REX_reg_breg(res, res), // movzbl
8412 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8413 ins_pipe( pipe_cmpxchg );
8414 %}
8416 //----------Subtraction Instructions-------------------------------------------
8418 // Integer Subtraction Instructions
8419 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8420 %{
8421 match(Set dst (SubI dst src));
8422 effect(KILL cr);
8424 format %{ "subl $dst, $src\t# int" %}
8425 opcode(0x2B);
8426 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
8427 ins_pipe(ialu_reg_reg);
8428 %}
8430 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
8431 %{
8432 match(Set dst (SubI dst src));
8433 effect(KILL cr);
8435 format %{ "subl $dst, $src\t# int" %}
8436 opcode(0x81, 0x05); /* Opcode 81 /5 */
8437 ins_encode(OpcSErm(dst, src), Con8or32(src));
8438 ins_pipe(ialu_reg);
8439 %}
8441 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
8442 %{
8443 match(Set dst (SubI dst (LoadI src)));
8444 effect(KILL cr);
8446 ins_cost(125);
8447 format %{ "subl $dst, $src\t# int" %}
8448 opcode(0x2B);
8449 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
8450 ins_pipe(ialu_reg_mem);
8451 %}
8453 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
8454 %{
8455 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8456 effect(KILL cr);
8458 ins_cost(150);
8459 format %{ "subl $dst, $src\t# int" %}
8460 opcode(0x29); /* Opcode 29 /r */
8461 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
8462 ins_pipe(ialu_mem_reg);
8463 %}
8465 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
8466 %{
8467 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8468 effect(KILL cr);
8470 ins_cost(125); // XXX
8471 format %{ "subl $dst, $src\t# int" %}
8472 opcode(0x81); /* Opcode 81 /5 id */
8473 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8474 ins_pipe(ialu_mem_imm);
8475 %}
8477 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8478 %{
8479 match(Set dst (SubL dst src));
8480 effect(KILL cr);
8482 format %{ "subq $dst, $src\t# long" %}
8483 opcode(0x2B);
8484 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8485 ins_pipe(ialu_reg_reg);
8486 %}
8488 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
8489 %{
8490 match(Set dst (SubL dst src));
8491 effect(KILL cr);
8493 format %{ "subq $dst, $src\t# long" %}
8494 opcode(0x81, 0x05); /* Opcode 81 /5 */
8495 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8496 ins_pipe(ialu_reg);
8497 %}
8499 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8500 %{
8501 match(Set dst (SubL dst (LoadL src)));
8502 effect(KILL cr);
8504 ins_cost(125);
8505 format %{ "subq $dst, $src\t# long" %}
8506 opcode(0x2B);
8507 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8508 ins_pipe(ialu_reg_mem);
8509 %}
8511 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8512 %{
8513 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8514 effect(KILL cr);
8516 ins_cost(150);
8517 format %{ "subq $dst, $src\t# long" %}
8518 opcode(0x29); /* Opcode 29 /r */
8519 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8520 ins_pipe(ialu_mem_reg);
8521 %}
8523 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8524 %{
8525 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8526 effect(KILL cr);
8528 ins_cost(125); // XXX
8529 format %{ "subq $dst, $src\t# long" %}
8530 opcode(0x81); /* Opcode 81 /5 id */
8531 ins_encode(REX_mem_wide(dst),
8532 OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8533 ins_pipe(ialu_mem_imm);
8534 %}
8536 // Subtract from a pointer
8537 // XXX hmpf???
8538 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
8539 %{
8540 match(Set dst (AddP dst (SubI zero src)));
8541 effect(KILL cr);
8543 format %{ "subq $dst, $src\t# ptr - int" %}
8544 opcode(0x2B);
8545 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8546 ins_pipe(ialu_reg_reg);
8547 %}
8549 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
8550 %{
8551 match(Set dst (SubI zero dst));
8552 effect(KILL cr);
8554 format %{ "negl $dst\t# int" %}
8555 opcode(0xF7, 0x03); // Opcode F7 /3
8556 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8557 ins_pipe(ialu_reg);
8558 %}
8560 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
8561 %{
8562 match(Set dst (StoreI dst (SubI zero (LoadI dst))));
8563 effect(KILL cr);
8565 format %{ "negl $dst\t# int" %}
8566 opcode(0xF7, 0x03); // Opcode F7 /3
8567 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8568 ins_pipe(ialu_reg);
8569 %}
8571 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
8572 %{
8573 match(Set dst (SubL zero dst));
8574 effect(KILL cr);
8576 format %{ "negq $dst\t# long" %}
8577 opcode(0xF7, 0x03); // Opcode F7 /3
8578 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8579 ins_pipe(ialu_reg);
8580 %}
8582 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
8583 %{
8584 match(Set dst (StoreL dst (SubL zero (LoadL dst))));
8585 effect(KILL cr);
8587 format %{ "negq $dst\t# long" %}
8588 opcode(0xF7, 0x03); // Opcode F7 /3
8589 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8590 ins_pipe(ialu_reg);
8591 %}
8594 //----------Multiplication/Division Instructions-------------------------------
8595 // Integer Multiplication Instructions
8596 // Multiply Register
8598 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8599 %{
8600 match(Set dst (MulI dst src));
8601 effect(KILL cr);
8603 ins_cost(300);
8604 format %{ "imull $dst, $src\t# int" %}
8605 opcode(0x0F, 0xAF);
8606 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8607 ins_pipe(ialu_reg_reg_alu0);
8608 %}
8610 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
8611 %{
8612 match(Set dst (MulI src imm));
8613 effect(KILL cr);
8615 ins_cost(300);
8616 format %{ "imull $dst, $src, $imm\t# int" %}
8617 opcode(0x69); /* 69 /r id */
8618 ins_encode(REX_reg_reg(dst, src),
8619 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8620 ins_pipe(ialu_reg_reg_alu0);
8621 %}
8623 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
8624 %{
8625 match(Set dst (MulI dst (LoadI src)));
8626 effect(KILL cr);
8628 ins_cost(350);
8629 format %{ "imull $dst, $src\t# int" %}
8630 opcode(0x0F, 0xAF);
8631 ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
8632 ins_pipe(ialu_reg_mem_alu0);
8633 %}
8635 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
8636 %{
8637 match(Set dst (MulI (LoadI src) imm));
8638 effect(KILL cr);
8640 ins_cost(300);
8641 format %{ "imull $dst, $src, $imm\t# int" %}
8642 opcode(0x69); /* 69 /r id */
8643 ins_encode(REX_reg_mem(dst, src),
8644 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8645 ins_pipe(ialu_reg_mem_alu0);
8646 %}
8648 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8649 %{
8650 match(Set dst (MulL dst src));
8651 effect(KILL cr);
8653 ins_cost(300);
8654 format %{ "imulq $dst, $src\t# long" %}
8655 opcode(0x0F, 0xAF);
8656 ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
8657 ins_pipe(ialu_reg_reg_alu0);
8658 %}
8660 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
8661 %{
8662 match(Set dst (MulL src imm));
8663 effect(KILL cr);
8665 ins_cost(300);
8666 format %{ "imulq $dst, $src, $imm\t# long" %}
8667 opcode(0x69); /* 69 /r id */
8668 ins_encode(REX_reg_reg_wide(dst, src),
8669 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8670 ins_pipe(ialu_reg_reg_alu0);
8671 %}
8673 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
8674 %{
8675 match(Set dst (MulL dst (LoadL src)));
8676 effect(KILL cr);
8678 ins_cost(350);
8679 format %{ "imulq $dst, $src\t# long" %}
8680 opcode(0x0F, 0xAF);
8681 ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
8682 ins_pipe(ialu_reg_mem_alu0);
8683 %}
8685 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
8686 %{
8687 match(Set dst (MulL (LoadL src) imm));
8688 effect(KILL cr);
8690 ins_cost(300);
8691 format %{ "imulq $dst, $src, $imm\t# long" %}
8692 opcode(0x69); /* 69 /r id */
8693 ins_encode(REX_reg_mem_wide(dst, src),
8694 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8695 ins_pipe(ialu_reg_mem_alu0);
8696 %}
8698 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8699 %{
8700 match(Set dst (MulHiL src rax));
8701 effect(USE_KILL rax, KILL cr);
8703 ins_cost(300);
8704 format %{ "imulq RDX:RAX, RAX, $src\t# mulhi" %}
8705 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8706 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8707 ins_pipe(ialu_reg_reg_alu0);
8708 %}
8710 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8711 rFlagsReg cr)
8712 %{
8713 match(Set rax (DivI rax div));
8714 effect(KILL rdx, KILL cr);
8716 ins_cost(30*100+10*100); // XXX
8717 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8718 "jne,s normal\n\t"
8719 "xorl rdx, rdx\n\t"
8720 "cmpl $div, -1\n\t"
8721 "je,s done\n"
8722 "normal: cdql\n\t"
8723 "idivl $div\n"
8724 "done:" %}
8725 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8726 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8727 ins_pipe(ialu_reg_reg_alu0);
8728 %}
8730 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8731 rFlagsReg cr)
8732 %{
8733 match(Set rax (DivL rax div));
8734 effect(KILL rdx, KILL cr);
8736 ins_cost(30*100+10*100); // XXX
8737 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8738 "cmpq rax, rdx\n\t"
8739 "jne,s normal\n\t"
8740 "xorl rdx, rdx\n\t"
8741 "cmpq $div, -1\n\t"
8742 "je,s done\n"
8743 "normal: cdqq\n\t"
8744 "idivq $div\n"
8745 "done:" %}
8746 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8747 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8748 ins_pipe(ialu_reg_reg_alu0);
8749 %}
8751 // Integer DIVMOD with Register, both quotient and mod results
8752 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8753 rFlagsReg cr)
8754 %{
8755 match(DivModI rax div);
8756 effect(KILL cr);
8758 ins_cost(30*100+10*100); // XXX
8759 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8760 "jne,s normal\n\t"
8761 "xorl rdx, rdx\n\t"
8762 "cmpl $div, -1\n\t"
8763 "je,s done\n"
8764 "normal: cdql\n\t"
8765 "idivl $div\n"
8766 "done:" %}
8767 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8768 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8769 ins_pipe(pipe_slow);
8770 %}
8772 // Long DIVMOD with Register, both quotient and mod results
8773 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8774 rFlagsReg cr)
8775 %{
8776 match(DivModL rax div);
8777 effect(KILL cr);
8779 ins_cost(30*100+10*100); // XXX
8780 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8781 "cmpq rax, rdx\n\t"
8782 "jne,s normal\n\t"
8783 "xorl rdx, rdx\n\t"
8784 "cmpq $div, -1\n\t"
8785 "je,s done\n"
8786 "normal: cdqq\n\t"
8787 "idivq $div\n"
8788 "done:" %}
8789 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8790 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8791 ins_pipe(pipe_slow);
8792 %}
8794 //----------- DivL-By-Constant-Expansions--------------------------------------
8795 // DivI cases are handled by the compiler
8797 // Magic constant, reciprocal of 10
8798 instruct loadConL_0x6666666666666667(rRegL dst)
8799 %{
8800 effect(DEF dst);
8802 format %{ "movq $dst, #0x666666666666667\t# Used in div-by-10" %}
8803 ins_encode(load_immL(dst, 0x6666666666666667));
8804 ins_pipe(ialu_reg);
8805 %}
8807 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8808 %{
8809 effect(DEF dst, USE src, USE_KILL rax, KILL cr);
8811 format %{ "imulq rdx:rax, rax, $src\t# Used in div-by-10" %}
8812 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8813 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8814 ins_pipe(ialu_reg_reg_alu0);
8815 %}
8817 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
8818 %{
8819 effect(USE_DEF dst, KILL cr);
8821 format %{ "sarq $dst, #63\t# Used in div-by-10" %}
8822 opcode(0xC1, 0x7); /* C1 /7 ib */
8823 ins_encode(reg_opc_imm_wide(dst, 0x3F));
8824 ins_pipe(ialu_reg);
8825 %}
8827 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
8828 %{
8829 effect(USE_DEF dst, KILL cr);
8831 format %{ "sarq $dst, #2\t# Used in div-by-10" %}
8832 opcode(0xC1, 0x7); /* C1 /7 ib */
8833 ins_encode(reg_opc_imm_wide(dst, 0x2));
8834 ins_pipe(ialu_reg);
8835 %}
8837 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
8838 %{
8839 match(Set dst (DivL src div));
8841 ins_cost((5+8)*100);
8842 expand %{
8843 rax_RegL rax; // Killed temp
8844 rFlagsReg cr; // Killed
8845 loadConL_0x6666666666666667(rax); // movq rax, 0x6666666666666667
8846 mul_hi(dst, src, rax, cr); // mulq rdx:rax <= rax * $src
8847 sarL_rReg_63(src, cr); // sarq src, 63
8848 sarL_rReg_2(dst, cr); // sarq rdx, 2
8849 subL_rReg(dst, src, cr); // subl rdx, src
8850 %}
8851 %}
8853 //-----------------------------------------------------------------------------
8855 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
8856 rFlagsReg cr)
8857 %{
8858 match(Set rdx (ModI rax div));
8859 effect(KILL rax, KILL cr);
8861 ins_cost(300); // XXX
8862 format %{ "cmpl rax, 0x80000000\t# irem\n\t"
8863 "jne,s normal\n\t"
8864 "xorl rdx, rdx\n\t"
8865 "cmpl $div, -1\n\t"
8866 "je,s done\n"
8867 "normal: cdql\n\t"
8868 "idivl $div\n"
8869 "done:" %}
8870 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8871 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8872 ins_pipe(ialu_reg_reg_alu0);
8873 %}
8875 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
8876 rFlagsReg cr)
8877 %{
8878 match(Set rdx (ModL rax div));
8879 effect(KILL rax, KILL cr);
8881 ins_cost(300); // XXX
8882 format %{ "movq rdx, 0x8000000000000000\t# lrem\n\t"
8883 "cmpq rax, rdx\n\t"
8884 "jne,s normal\n\t"
8885 "xorl rdx, rdx\n\t"
8886 "cmpq $div, -1\n\t"
8887 "je,s done\n"
8888 "normal: cdqq\n\t"
8889 "idivq $div\n"
8890 "done:" %}
8891 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8892 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8893 ins_pipe(ialu_reg_reg_alu0);
8894 %}
8896 // Integer Shift Instructions
8897 // Shift Left by one
8898 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8899 %{
8900 match(Set dst (LShiftI dst shift));
8901 effect(KILL cr);
8903 format %{ "sall $dst, $shift" %}
8904 opcode(0xD1, 0x4); /* D1 /4 */
8905 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8906 ins_pipe(ialu_reg);
8907 %}
8909 // Shift Left by one
8910 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8911 %{
8912 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8913 effect(KILL cr);
8915 format %{ "sall $dst, $shift\t" %}
8916 opcode(0xD1, 0x4); /* D1 /4 */
8917 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8918 ins_pipe(ialu_mem_imm);
8919 %}
8921 // Shift Left by 8-bit immediate
8922 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8923 %{
8924 match(Set dst (LShiftI dst shift));
8925 effect(KILL cr);
8927 format %{ "sall $dst, $shift" %}
8928 opcode(0xC1, 0x4); /* C1 /4 ib */
8929 ins_encode(reg_opc_imm(dst, shift));
8930 ins_pipe(ialu_reg);
8931 %}
8933 // Shift Left by 8-bit immediate
8934 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
8935 %{
8936 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8937 effect(KILL cr);
8939 format %{ "sall $dst, $shift" %}
8940 opcode(0xC1, 0x4); /* C1 /4 ib */
8941 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
8942 ins_pipe(ialu_mem_imm);
8943 %}
8945 // Shift Left by variable
8946 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
8947 %{
8948 match(Set dst (LShiftI dst shift));
8949 effect(KILL cr);
8951 format %{ "sall $dst, $shift" %}
8952 opcode(0xD3, 0x4); /* D3 /4 */
8953 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8954 ins_pipe(ialu_reg_reg);
8955 %}
8957 // Shift Left by variable
8958 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
8959 %{
8960 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
8961 effect(KILL cr);
8963 format %{ "sall $dst, $shift" %}
8964 opcode(0xD3, 0x4); /* D3 /4 */
8965 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8966 ins_pipe(ialu_mem_reg);
8967 %}
8969 // Arithmetic shift right by one
8970 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
8971 %{
8972 match(Set dst (RShiftI dst shift));
8973 effect(KILL cr);
8975 format %{ "sarl $dst, $shift" %}
8976 opcode(0xD1, 0x7); /* D1 /7 */
8977 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8978 ins_pipe(ialu_reg);
8979 %}
8981 // Arithmetic shift right by one
8982 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
8983 %{
8984 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
8985 effect(KILL cr);
8987 format %{ "sarl $dst, $shift" %}
8988 opcode(0xD1, 0x7); /* D1 /7 */
8989 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8990 ins_pipe(ialu_mem_imm);
8991 %}
8993 // Arithmetic Shift Right by 8-bit immediate
8994 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
8995 %{
8996 match(Set dst (RShiftI dst shift));
8997 effect(KILL cr);
8999 format %{ "sarl $dst, $shift" %}
9000 opcode(0xC1, 0x7); /* C1 /7 ib */
9001 ins_encode(reg_opc_imm(dst, shift));
9002 ins_pipe(ialu_mem_imm);
9003 %}
9005 // Arithmetic Shift Right by 8-bit immediate
9006 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9007 %{
9008 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
9009 effect(KILL cr);
9011 format %{ "sarl $dst, $shift" %}
9012 opcode(0xC1, 0x7); /* C1 /7 ib */
9013 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9014 ins_pipe(ialu_mem_imm);
9015 %}
9017 // Arithmetic Shift Right by variable
9018 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9019 %{
9020 match(Set dst (RShiftI dst shift));
9021 effect(KILL cr);
9023 format %{ "sarl $dst, $shift" %}
9024 opcode(0xD3, 0x7); /* D3 /7 */
9025 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9026 ins_pipe(ialu_reg_reg);
9027 %}
9029 // Arithmetic Shift Right by variable
9030 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9031 %{
9032 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
9033 effect(KILL cr);
9035 format %{ "sarl $dst, $shift" %}
9036 opcode(0xD3, 0x7); /* D3 /7 */
9037 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9038 ins_pipe(ialu_mem_reg);
9039 %}
9041 // Logical shift right by one
9042 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
9043 %{
9044 match(Set dst (URShiftI dst shift));
9045 effect(KILL cr);
9047 format %{ "shrl $dst, $shift" %}
9048 opcode(0xD1, 0x5); /* D1 /5 */
9049 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9050 ins_pipe(ialu_reg);
9051 %}
9053 // Logical shift right by one
9054 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9055 %{
9056 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9057 effect(KILL cr);
9059 format %{ "shrl $dst, $shift" %}
9060 opcode(0xD1, 0x5); /* D1 /5 */
9061 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9062 ins_pipe(ialu_mem_imm);
9063 %}
9065 // Logical Shift Right by 8-bit immediate
9066 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
9067 %{
9068 match(Set dst (URShiftI dst shift));
9069 effect(KILL cr);
9071 format %{ "shrl $dst, $shift" %}
9072 opcode(0xC1, 0x5); /* C1 /5 ib */
9073 ins_encode(reg_opc_imm(dst, shift));
9074 ins_pipe(ialu_reg);
9075 %}
9077 // Logical Shift Right by 8-bit immediate
9078 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9079 %{
9080 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9081 effect(KILL cr);
9083 format %{ "shrl $dst, $shift" %}
9084 opcode(0xC1, 0x5); /* C1 /5 ib */
9085 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9086 ins_pipe(ialu_mem_imm);
9087 %}
9089 // Logical Shift Right by variable
9090 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9091 %{
9092 match(Set dst (URShiftI dst shift));
9093 effect(KILL cr);
9095 format %{ "shrl $dst, $shift" %}
9096 opcode(0xD3, 0x5); /* D3 /5 */
9097 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9098 ins_pipe(ialu_reg_reg);
9099 %}
9101 // Logical Shift Right by variable
9102 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9103 %{
9104 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9105 effect(KILL cr);
9107 format %{ "shrl $dst, $shift" %}
9108 opcode(0xD3, 0x5); /* D3 /5 */
9109 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9110 ins_pipe(ialu_mem_reg);
9111 %}
9113 // Long Shift Instructions
9114 // Shift Left by one
9115 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9116 %{
9117 match(Set dst (LShiftL dst shift));
9118 effect(KILL cr);
9120 format %{ "salq $dst, $shift" %}
9121 opcode(0xD1, 0x4); /* D1 /4 */
9122 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9123 ins_pipe(ialu_reg);
9124 %}
9126 // Shift Left by one
9127 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9128 %{
9129 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9130 effect(KILL cr);
9132 format %{ "salq $dst, $shift" %}
9133 opcode(0xD1, 0x4); /* D1 /4 */
9134 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9135 ins_pipe(ialu_mem_imm);
9136 %}
9138 // Shift Left by 8-bit immediate
9139 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9140 %{
9141 match(Set dst (LShiftL dst shift));
9142 effect(KILL cr);
9144 format %{ "salq $dst, $shift" %}
9145 opcode(0xC1, 0x4); /* C1 /4 ib */
9146 ins_encode(reg_opc_imm_wide(dst, shift));
9147 ins_pipe(ialu_reg);
9148 %}
9150 // Shift Left by 8-bit immediate
9151 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9152 %{
9153 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9154 effect(KILL cr);
9156 format %{ "salq $dst, $shift" %}
9157 opcode(0xC1, 0x4); /* C1 /4 ib */
9158 ins_encode(REX_mem_wide(dst), OpcP,
9159 RM_opc_mem(secondary, dst), Con8or32(shift));
9160 ins_pipe(ialu_mem_imm);
9161 %}
9163 // Shift Left by variable
9164 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9165 %{
9166 match(Set dst (LShiftL dst shift));
9167 effect(KILL cr);
9169 format %{ "salq $dst, $shift" %}
9170 opcode(0xD3, 0x4); /* D3 /4 */
9171 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9172 ins_pipe(ialu_reg_reg);
9173 %}
9175 // Shift Left by variable
9176 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9177 %{
9178 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9179 effect(KILL cr);
9181 format %{ "salq $dst, $shift" %}
9182 opcode(0xD3, 0x4); /* D3 /4 */
9183 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9184 ins_pipe(ialu_mem_reg);
9185 %}
9187 // Arithmetic shift right by one
9188 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9189 %{
9190 match(Set dst (RShiftL dst shift));
9191 effect(KILL cr);
9193 format %{ "sarq $dst, $shift" %}
9194 opcode(0xD1, 0x7); /* D1 /7 */
9195 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9196 ins_pipe(ialu_reg);
9197 %}
9199 // Arithmetic shift right by one
9200 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9201 %{
9202 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9203 effect(KILL cr);
9205 format %{ "sarq $dst, $shift" %}
9206 opcode(0xD1, 0x7); /* D1 /7 */
9207 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9208 ins_pipe(ialu_mem_imm);
9209 %}
9211 // Arithmetic Shift Right by 8-bit immediate
9212 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9213 %{
9214 match(Set dst (RShiftL dst shift));
9215 effect(KILL cr);
9217 format %{ "sarq $dst, $shift" %}
9218 opcode(0xC1, 0x7); /* C1 /7 ib */
9219 ins_encode(reg_opc_imm_wide(dst, shift));
9220 ins_pipe(ialu_mem_imm);
9221 %}
9223 // Arithmetic Shift Right by 8-bit immediate
9224 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9225 %{
9226 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9227 effect(KILL cr);
9229 format %{ "sarq $dst, $shift" %}
9230 opcode(0xC1, 0x7); /* C1 /7 ib */
9231 ins_encode(REX_mem_wide(dst), OpcP,
9232 RM_opc_mem(secondary, dst), Con8or32(shift));
9233 ins_pipe(ialu_mem_imm);
9234 %}
9236 // Arithmetic Shift Right by variable
9237 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9238 %{
9239 match(Set dst (RShiftL dst shift));
9240 effect(KILL cr);
9242 format %{ "sarq $dst, $shift" %}
9243 opcode(0xD3, 0x7); /* D3 /7 */
9244 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9245 ins_pipe(ialu_reg_reg);
9246 %}
9248 // Arithmetic Shift Right by variable
9249 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9250 %{
9251 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9252 effect(KILL cr);
9254 format %{ "sarq $dst, $shift" %}
9255 opcode(0xD3, 0x7); /* D3 /7 */
9256 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9257 ins_pipe(ialu_mem_reg);
9258 %}
9260 // Logical shift right by one
9261 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9262 %{
9263 match(Set dst (URShiftL dst shift));
9264 effect(KILL cr);
9266 format %{ "shrq $dst, $shift" %}
9267 opcode(0xD1, 0x5); /* D1 /5 */
9268 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
9269 ins_pipe(ialu_reg);
9270 %}
9272 // Logical shift right by one
9273 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9274 %{
9275 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9276 effect(KILL cr);
9278 format %{ "shrq $dst, $shift" %}
9279 opcode(0xD1, 0x5); /* D1 /5 */
9280 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9281 ins_pipe(ialu_mem_imm);
9282 %}
9284 // Logical Shift Right by 8-bit immediate
9285 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9286 %{
9287 match(Set dst (URShiftL dst shift));
9288 effect(KILL cr);
9290 format %{ "shrq $dst, $shift" %}
9291 opcode(0xC1, 0x5); /* C1 /5 ib */
9292 ins_encode(reg_opc_imm_wide(dst, shift));
9293 ins_pipe(ialu_reg);
9294 %}
9297 // Logical Shift Right by 8-bit immediate
9298 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9299 %{
9300 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9301 effect(KILL cr);
9303 format %{ "shrq $dst, $shift" %}
9304 opcode(0xC1, 0x5); /* C1 /5 ib */
9305 ins_encode(REX_mem_wide(dst), OpcP,
9306 RM_opc_mem(secondary, dst), Con8or32(shift));
9307 ins_pipe(ialu_mem_imm);
9308 %}
9310 // Logical Shift Right by variable
9311 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9312 %{
9313 match(Set dst (URShiftL dst shift));
9314 effect(KILL cr);
9316 format %{ "shrq $dst, $shift" %}
9317 opcode(0xD3, 0x5); /* D3 /5 */
9318 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9319 ins_pipe(ialu_reg_reg);
9320 %}
9322 // Logical Shift Right by variable
9323 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9324 %{
9325 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9326 effect(KILL cr);
9328 format %{ "shrq $dst, $shift" %}
9329 opcode(0xD3, 0x5); /* D3 /5 */
9330 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9331 ins_pipe(ialu_mem_reg);
9332 %}
9334 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
9335 // This idiom is used by the compiler for the i2b bytecode.
9336 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
9337 %{
9338 match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
9340 format %{ "movsbl $dst, $src\t# i2b" %}
9341 opcode(0x0F, 0xBE);
9342 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9343 ins_pipe(ialu_reg_reg);
9344 %}
9346 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
9347 // This idiom is used by the compiler the i2s bytecode.
9348 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
9349 %{
9350 match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
9352 format %{ "movswl $dst, $src\t# i2s" %}
9353 opcode(0x0F, 0xBF);
9354 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9355 ins_pipe(ialu_reg_reg);
9356 %}
9358 // ROL/ROR instructions
9360 // ROL expand
9361 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
9362 effect(KILL cr, USE_DEF dst);
9364 format %{ "roll $dst" %}
9365 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9366 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9367 ins_pipe(ialu_reg);
9368 %}
9370 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
9371 effect(USE_DEF dst, USE shift, KILL cr);
9373 format %{ "roll $dst, $shift" %}
9374 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9375 ins_encode( reg_opc_imm(dst, shift) );
9376 ins_pipe(ialu_reg);
9377 %}
9379 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9380 %{
9381 effect(USE_DEF dst, USE shift, KILL cr);
9383 format %{ "roll $dst, $shift" %}
9384 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9385 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9386 ins_pipe(ialu_reg_reg);
9387 %}
9388 // end of ROL expand
9390 // Rotate Left by one
9391 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9392 %{
9393 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9395 expand %{
9396 rolI_rReg_imm1(dst, cr);
9397 %}
9398 %}
9400 // Rotate Left by 8-bit immediate
9401 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9402 %{
9403 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9404 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9406 expand %{
9407 rolI_rReg_imm8(dst, lshift, cr);
9408 %}
9409 %}
9411 // Rotate Left by variable
9412 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9413 %{
9414 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
9416 expand %{
9417 rolI_rReg_CL(dst, shift, cr);
9418 %}
9419 %}
9421 // Rotate Left by variable
9422 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9423 %{
9424 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
9426 expand %{
9427 rolI_rReg_CL(dst, shift, cr);
9428 %}
9429 %}
9431 // ROR expand
9432 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
9433 %{
9434 effect(USE_DEF dst, KILL cr);
9436 format %{ "rorl $dst" %}
9437 opcode(0xD1, 0x1); /* D1 /1 */
9438 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9439 ins_pipe(ialu_reg);
9440 %}
9442 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
9443 %{
9444 effect(USE_DEF dst, USE shift, KILL cr);
9446 format %{ "rorl $dst, $shift" %}
9447 opcode(0xC1, 0x1); /* C1 /1 ib */
9448 ins_encode(reg_opc_imm(dst, shift));
9449 ins_pipe(ialu_reg);
9450 %}
9452 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9453 %{
9454 effect(USE_DEF dst, USE shift, KILL cr);
9456 format %{ "rorl $dst, $shift" %}
9457 opcode(0xD3, 0x1); /* D3 /1 */
9458 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9459 ins_pipe(ialu_reg_reg);
9460 %}
9461 // end of ROR expand
9463 // Rotate Right by one
9464 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9465 %{
9466 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9468 expand %{
9469 rorI_rReg_imm1(dst, cr);
9470 %}
9471 %}
9473 // Rotate Right by 8-bit immediate
9474 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9475 %{
9476 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9477 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9479 expand %{
9480 rorI_rReg_imm8(dst, rshift, cr);
9481 %}
9482 %}
9484 // Rotate Right by variable
9485 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9486 %{
9487 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
9489 expand %{
9490 rorI_rReg_CL(dst, shift, cr);
9491 %}
9492 %}
9494 // Rotate Right by variable
9495 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9496 %{
9497 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
9499 expand %{
9500 rorI_rReg_CL(dst, shift, cr);
9501 %}
9502 %}
9504 // for long rotate
9505 // ROL expand
9506 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
9507 effect(USE_DEF dst, KILL cr);
9509 format %{ "rolq $dst" %}
9510 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9511 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9512 ins_pipe(ialu_reg);
9513 %}
9515 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
9516 effect(USE_DEF dst, USE shift, KILL cr);
9518 format %{ "rolq $dst, $shift" %}
9519 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9520 ins_encode( reg_opc_imm_wide(dst, shift) );
9521 ins_pipe(ialu_reg);
9522 %}
9524 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9525 %{
9526 effect(USE_DEF dst, USE shift, KILL cr);
9528 format %{ "rolq $dst, $shift" %}
9529 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9530 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9531 ins_pipe(ialu_reg_reg);
9532 %}
9533 // end of ROL expand
9535 // Rotate Left by one
9536 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9537 %{
9538 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9540 expand %{
9541 rolL_rReg_imm1(dst, cr);
9542 %}
9543 %}
9545 // Rotate Left by 8-bit immediate
9546 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9547 %{
9548 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9549 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9551 expand %{
9552 rolL_rReg_imm8(dst, lshift, cr);
9553 %}
9554 %}
9556 // Rotate Left by variable
9557 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9558 %{
9559 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
9561 expand %{
9562 rolL_rReg_CL(dst, shift, cr);
9563 %}
9564 %}
9566 // Rotate Left by variable
9567 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9568 %{
9569 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
9571 expand %{
9572 rolL_rReg_CL(dst, shift, cr);
9573 %}
9574 %}
9576 // ROR expand
9577 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
9578 %{
9579 effect(USE_DEF dst, KILL cr);
9581 format %{ "rorq $dst" %}
9582 opcode(0xD1, 0x1); /* D1 /1 */
9583 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9584 ins_pipe(ialu_reg);
9585 %}
9587 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
9588 %{
9589 effect(USE_DEF dst, USE shift, KILL cr);
9591 format %{ "rorq $dst, $shift" %}
9592 opcode(0xC1, 0x1); /* C1 /1 ib */
9593 ins_encode(reg_opc_imm_wide(dst, shift));
9594 ins_pipe(ialu_reg);
9595 %}
9597 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9598 %{
9599 effect(USE_DEF dst, USE shift, KILL cr);
9601 format %{ "rorq $dst, $shift" %}
9602 opcode(0xD3, 0x1); /* D3 /1 */
9603 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9604 ins_pipe(ialu_reg_reg);
9605 %}
9606 // end of ROR expand
9608 // Rotate Right by one
9609 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9610 %{
9611 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9613 expand %{
9614 rorL_rReg_imm1(dst, cr);
9615 %}
9616 %}
9618 // Rotate Right by 8-bit immediate
9619 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9620 %{
9621 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9622 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9624 expand %{
9625 rorL_rReg_imm8(dst, rshift, cr);
9626 %}
9627 %}
9629 // Rotate Right by variable
9630 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9631 %{
9632 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
9634 expand %{
9635 rorL_rReg_CL(dst, shift, cr);
9636 %}
9637 %}
9639 // Rotate Right by variable
9640 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9641 %{
9642 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
9644 expand %{
9645 rorL_rReg_CL(dst, shift, cr);
9646 %}
9647 %}
9649 // Logical Instructions
9651 // Integer Logical Instructions
9653 // And Instructions
9654 // And Register with Register
9655 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9656 %{
9657 match(Set dst (AndI dst src));
9658 effect(KILL cr);
9660 format %{ "andl $dst, $src\t# int" %}
9661 opcode(0x23);
9662 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9663 ins_pipe(ialu_reg_reg);
9664 %}
9666 // And Register with Immediate 255
9667 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
9668 %{
9669 match(Set dst (AndI dst src));
9671 format %{ "movzbl $dst, $dst\t# int & 0xFF" %}
9672 opcode(0x0F, 0xB6);
9673 ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9674 ins_pipe(ialu_reg);
9675 %}
9677 // And Register with Immediate 255 and promote to long
9678 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
9679 %{
9680 match(Set dst (ConvI2L (AndI src mask)));
9682 format %{ "movzbl $dst, $src\t# int & 0xFF -> long" %}
9683 opcode(0x0F, 0xB6);
9684 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9685 ins_pipe(ialu_reg);
9686 %}
9688 // And Register with Immediate 65535
9689 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
9690 %{
9691 match(Set dst (AndI dst src));
9693 format %{ "movzwl $dst, $dst\t# int & 0xFFFF" %}
9694 opcode(0x0F, 0xB7);
9695 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9696 ins_pipe(ialu_reg);
9697 %}
9699 // And Register with Immediate 65535 and promote to long
9700 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
9701 %{
9702 match(Set dst (ConvI2L (AndI src mask)));
9704 format %{ "movzwl $dst, $src\t# int & 0xFFFF -> long" %}
9705 opcode(0x0F, 0xB7);
9706 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9707 ins_pipe(ialu_reg);
9708 %}
9710 // And Register with Immediate
9711 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9712 %{
9713 match(Set dst (AndI dst src));
9714 effect(KILL cr);
9716 format %{ "andl $dst, $src\t# int" %}
9717 opcode(0x81, 0x04); /* Opcode 81 /4 */
9718 ins_encode(OpcSErm(dst, src), Con8or32(src));
9719 ins_pipe(ialu_reg);
9720 %}
9722 // And Register with Memory
9723 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9724 %{
9725 match(Set dst (AndI dst (LoadI src)));
9726 effect(KILL cr);
9728 ins_cost(125);
9729 format %{ "andl $dst, $src\t# int" %}
9730 opcode(0x23);
9731 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9732 ins_pipe(ialu_reg_mem);
9733 %}
9735 // And Memory with Register
9736 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9737 %{
9738 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9739 effect(KILL cr);
9741 ins_cost(150);
9742 format %{ "andl $dst, $src\t# int" %}
9743 opcode(0x21); /* Opcode 21 /r */
9744 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9745 ins_pipe(ialu_mem_reg);
9746 %}
9748 // And Memory with Immediate
9749 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
9750 %{
9751 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9752 effect(KILL cr);
9754 ins_cost(125);
9755 format %{ "andl $dst, $src\t# int" %}
9756 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9757 ins_encode(REX_mem(dst), OpcSE(src),
9758 RM_opc_mem(secondary, dst), Con8or32(src));
9759 ins_pipe(ialu_mem_imm);
9760 %}
9762 // Or Instructions
9763 // Or Register with Register
9764 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9765 %{
9766 match(Set dst (OrI dst src));
9767 effect(KILL cr);
9769 format %{ "orl $dst, $src\t# int" %}
9770 opcode(0x0B);
9771 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9772 ins_pipe(ialu_reg_reg);
9773 %}
9775 // Or Register with Immediate
9776 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9777 %{
9778 match(Set dst (OrI dst src));
9779 effect(KILL cr);
9781 format %{ "orl $dst, $src\t# int" %}
9782 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9783 ins_encode(OpcSErm(dst, src), Con8or32(src));
9784 ins_pipe(ialu_reg);
9785 %}
9787 // Or Register with Memory
9788 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9789 %{
9790 match(Set dst (OrI dst (LoadI src)));
9791 effect(KILL cr);
9793 ins_cost(125);
9794 format %{ "orl $dst, $src\t# int" %}
9795 opcode(0x0B);
9796 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9797 ins_pipe(ialu_reg_mem);
9798 %}
9800 // Or Memory with Register
9801 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9802 %{
9803 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9804 effect(KILL cr);
9806 ins_cost(150);
9807 format %{ "orl $dst, $src\t# int" %}
9808 opcode(0x09); /* Opcode 09 /r */
9809 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9810 ins_pipe(ialu_mem_reg);
9811 %}
9813 // Or Memory with Immediate
9814 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
9815 %{
9816 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9817 effect(KILL cr);
9819 ins_cost(125);
9820 format %{ "orl $dst, $src\t# int" %}
9821 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9822 ins_encode(REX_mem(dst), OpcSE(src),
9823 RM_opc_mem(secondary, dst), Con8or32(src));
9824 ins_pipe(ialu_mem_imm);
9825 %}
9827 // Xor Instructions
9828 // Xor Register with Register
9829 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9830 %{
9831 match(Set dst (XorI dst src));
9832 effect(KILL cr);
9834 format %{ "xorl $dst, $src\t# int" %}
9835 opcode(0x33);
9836 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9837 ins_pipe(ialu_reg_reg);
9838 %}
9840 // Xor Register with Immediate -1
9841 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{
9842 match(Set dst (XorI dst imm));
9844 format %{ "not $dst" %}
9845 ins_encode %{
9846 __ notl($dst$$Register);
9847 %}
9848 ins_pipe(ialu_reg);
9849 %}
9851 // Xor Register with Immediate
9852 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9853 %{
9854 match(Set dst (XorI dst src));
9855 effect(KILL cr);
9857 format %{ "xorl $dst, $src\t# int" %}
9858 opcode(0x81, 0x06); /* Opcode 81 /6 id */
9859 ins_encode(OpcSErm(dst, src), Con8or32(src));
9860 ins_pipe(ialu_reg);
9861 %}
9863 // Xor Register with Memory
9864 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9865 %{
9866 match(Set dst (XorI dst (LoadI src)));
9867 effect(KILL cr);
9869 ins_cost(125);
9870 format %{ "xorl $dst, $src\t# int" %}
9871 opcode(0x33);
9872 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9873 ins_pipe(ialu_reg_mem);
9874 %}
9876 // Xor Memory with Register
9877 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9878 %{
9879 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9880 effect(KILL cr);
9882 ins_cost(150);
9883 format %{ "xorl $dst, $src\t# int" %}
9884 opcode(0x31); /* Opcode 31 /r */
9885 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9886 ins_pipe(ialu_mem_reg);
9887 %}
9889 // Xor Memory with Immediate
9890 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
9891 %{
9892 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
9893 effect(KILL cr);
9895 ins_cost(125);
9896 format %{ "xorl $dst, $src\t# int" %}
9897 opcode(0x81, 0x6); /* Opcode 81 /6 id */
9898 ins_encode(REX_mem(dst), OpcSE(src),
9899 RM_opc_mem(secondary, dst), Con8or32(src));
9900 ins_pipe(ialu_mem_imm);
9901 %}
9904 // Long Logical Instructions
9906 // And Instructions
9907 // And Register with Register
9908 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9909 %{
9910 match(Set dst (AndL dst src));
9911 effect(KILL cr);
9913 format %{ "andq $dst, $src\t# long" %}
9914 opcode(0x23);
9915 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
9916 ins_pipe(ialu_reg_reg);
9917 %}
9919 // And Register with Immediate 255
9920 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
9921 %{
9922 match(Set dst (AndL dst src));
9924 format %{ "movzbq $dst, $dst\t# long & 0xFF" %}
9925 opcode(0x0F, 0xB6);
9926 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9927 ins_pipe(ialu_reg);
9928 %}
9930 // And Register with Immediate 65535
9931 instruct andL_rReg_imm65535(rRegL dst, immL_65535 src)
9932 %{
9933 match(Set dst (AndL dst src));
9935 format %{ "movzwq $dst, $dst\t# long & 0xFFFF" %}
9936 opcode(0x0F, 0xB7);
9937 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9938 ins_pipe(ialu_reg);
9939 %}
9941 // And Register with Immediate
9942 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
9943 %{
9944 match(Set dst (AndL dst src));
9945 effect(KILL cr);
9947 format %{ "andq $dst, $src\t# long" %}
9948 opcode(0x81, 0x04); /* Opcode 81 /4 */
9949 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
9950 ins_pipe(ialu_reg);
9951 %}
9953 // And Register with Memory
9954 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
9955 %{
9956 match(Set dst (AndL dst (LoadL src)));
9957 effect(KILL cr);
9959 ins_cost(125);
9960 format %{ "andq $dst, $src\t# long" %}
9961 opcode(0x23);
9962 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
9963 ins_pipe(ialu_reg_mem);
9964 %}
9966 // And Memory with Register
9967 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
9968 %{
9969 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9970 effect(KILL cr);
9972 ins_cost(150);
9973 format %{ "andq $dst, $src\t# long" %}
9974 opcode(0x21); /* Opcode 21 /r */
9975 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
9976 ins_pipe(ialu_mem_reg);
9977 %}
9979 // And Memory with Immediate
9980 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
9981 %{
9982 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
9983 effect(KILL cr);
9985 ins_cost(125);
9986 format %{ "andq $dst, $src\t# long" %}
9987 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9988 ins_encode(REX_mem_wide(dst), OpcSE(src),
9989 RM_opc_mem(secondary, dst), Con8or32(src));
9990 ins_pipe(ialu_mem_imm);
9991 %}
9993 // Or Instructions
9994 // Or Register with Register
9995 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
9996 %{
9997 match(Set dst (OrL dst src));
9998 effect(KILL cr);
10000 format %{ "orq $dst, $src\t# long" %}
10001 opcode(0x0B);
10002 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10003 ins_pipe(ialu_reg_reg);
10004 %}
10006 // Use any_RegP to match R15 (TLS register) without spilling.
10007 instruct orL_rReg_castP2X(rRegL dst, any_RegP src, rFlagsReg cr) %{
10008 match(Set dst (OrL dst (CastP2X src)));
10009 effect(KILL cr);
10011 format %{ "orq $dst, $src\t# long" %}
10012 opcode(0x0B);
10013 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10014 ins_pipe(ialu_reg_reg);
10015 %}
10018 // Or Register with Immediate
10019 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10020 %{
10021 match(Set dst (OrL dst src));
10022 effect(KILL cr);
10024 format %{ "orq $dst, $src\t# long" %}
10025 opcode(0x81, 0x01); /* Opcode 81 /1 id */
10026 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10027 ins_pipe(ialu_reg);
10028 %}
10030 // Or Register with Memory
10031 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10032 %{
10033 match(Set dst (OrL dst (LoadL src)));
10034 effect(KILL cr);
10036 ins_cost(125);
10037 format %{ "orq $dst, $src\t# long" %}
10038 opcode(0x0B);
10039 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10040 ins_pipe(ialu_reg_mem);
10041 %}
10043 // Or Memory with Register
10044 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10045 %{
10046 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
10047 effect(KILL cr);
10049 ins_cost(150);
10050 format %{ "orq $dst, $src\t# long" %}
10051 opcode(0x09); /* Opcode 09 /r */
10052 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10053 ins_pipe(ialu_mem_reg);
10054 %}
10056 // Or Memory with Immediate
10057 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10058 %{
10059 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
10060 effect(KILL cr);
10062 ins_cost(125);
10063 format %{ "orq $dst, $src\t# long" %}
10064 opcode(0x81, 0x1); /* Opcode 81 /1 id */
10065 ins_encode(REX_mem_wide(dst), OpcSE(src),
10066 RM_opc_mem(secondary, dst), Con8or32(src));
10067 ins_pipe(ialu_mem_imm);
10068 %}
10070 // Xor Instructions
10071 // Xor Register with Register
10072 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
10073 %{
10074 match(Set dst (XorL dst src));
10075 effect(KILL cr);
10077 format %{ "xorq $dst, $src\t# long" %}
10078 opcode(0x33);
10079 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10080 ins_pipe(ialu_reg_reg);
10081 %}
10083 // Xor Register with Immediate -1
10084 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{
10085 match(Set dst (XorL dst imm));
10087 format %{ "notq $dst" %}
10088 ins_encode %{
10089 __ notq($dst$$Register);
10090 %}
10091 ins_pipe(ialu_reg);
10092 %}
10094 // Xor Register with Immediate
10095 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10096 %{
10097 match(Set dst (XorL dst src));
10098 effect(KILL cr);
10100 format %{ "xorq $dst, $src\t# long" %}
10101 opcode(0x81, 0x06); /* Opcode 81 /6 id */
10102 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10103 ins_pipe(ialu_reg);
10104 %}
10106 // Xor Register with Memory
10107 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10108 %{
10109 match(Set dst (XorL dst (LoadL src)));
10110 effect(KILL cr);
10112 ins_cost(125);
10113 format %{ "xorq $dst, $src\t# long" %}
10114 opcode(0x33);
10115 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10116 ins_pipe(ialu_reg_mem);
10117 %}
10119 // Xor Memory with Register
10120 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10121 %{
10122 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10123 effect(KILL cr);
10125 ins_cost(150);
10126 format %{ "xorq $dst, $src\t# long" %}
10127 opcode(0x31); /* Opcode 31 /r */
10128 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10129 ins_pipe(ialu_mem_reg);
10130 %}
10132 // Xor Memory with Immediate
10133 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10134 %{
10135 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10136 effect(KILL cr);
10138 ins_cost(125);
10139 format %{ "xorq $dst, $src\t# long" %}
10140 opcode(0x81, 0x6); /* Opcode 81 /6 id */
10141 ins_encode(REX_mem_wide(dst), OpcSE(src),
10142 RM_opc_mem(secondary, dst), Con8or32(src));
10143 ins_pipe(ialu_mem_imm);
10144 %}
10146 // Convert Int to Boolean
10147 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
10148 %{
10149 match(Set dst (Conv2B src));
10150 effect(KILL cr);
10152 format %{ "testl $src, $src\t# ci2b\n\t"
10153 "setnz $dst\n\t"
10154 "movzbl $dst, $dst" %}
10155 ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
10156 setNZ_reg(dst),
10157 REX_reg_breg(dst, dst), // movzbl
10158 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10159 ins_pipe(pipe_slow); // XXX
10160 %}
10162 // Convert Pointer to Boolean
10163 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
10164 %{
10165 match(Set dst (Conv2B src));
10166 effect(KILL cr);
10168 format %{ "testq $src, $src\t# cp2b\n\t"
10169 "setnz $dst\n\t"
10170 "movzbl $dst, $dst" %}
10171 ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
10172 setNZ_reg(dst),
10173 REX_reg_breg(dst, dst), // movzbl
10174 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10175 ins_pipe(pipe_slow); // XXX
10176 %}
10178 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
10179 %{
10180 match(Set dst (CmpLTMask p q));
10181 effect(KILL cr);
10183 ins_cost(400); // XXX
10184 format %{ "cmpl $p, $q\t# cmpLTMask\n\t"
10185 "setlt $dst\n\t"
10186 "movzbl $dst, $dst\n\t"
10187 "negl $dst" %}
10188 ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
10189 setLT_reg(dst),
10190 REX_reg_breg(dst, dst), // movzbl
10191 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
10192 neg_reg(dst));
10193 ins_pipe(pipe_slow);
10194 %}
10196 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
10197 %{
10198 match(Set dst (CmpLTMask dst zero));
10199 effect(KILL cr);
10201 ins_cost(100); // XXX
10202 format %{ "sarl $dst, #31\t# cmpLTMask0" %}
10203 opcode(0xC1, 0x7); /* C1 /7 ib */
10204 ins_encode(reg_opc_imm(dst, 0x1F));
10205 ins_pipe(ialu_reg);
10206 %}
10209 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y,
10210 rRegI tmp,
10211 rFlagsReg cr)
10212 %{
10213 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
10214 effect(TEMP tmp, KILL cr);
10216 ins_cost(400); // XXX
10217 format %{ "subl $p, $q\t# cadd_cmpLTMask1\n\t"
10218 "sbbl $tmp, $tmp\n\t"
10219 "andl $tmp, $y\n\t"
10220 "addl $p, $tmp" %}
10221 ins_encode(enc_cmpLTP(p, q, y, tmp));
10222 ins_pipe(pipe_cmplt);
10223 %}
10225 /* If I enable this, I encourage spilling in the inner loop of compress.
10226 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr )
10227 %{
10228 match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
10229 effect( TEMP tmp, KILL cr );
10230 ins_cost(400);
10232 format %{ "SUB $p,$q\n\t"
10233 "SBB RCX,RCX\n\t"
10234 "AND RCX,$y\n\t"
10235 "ADD $p,RCX" %}
10236 ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
10237 %}
10238 */
10240 //---------- FP Instructions------------------------------------------------
10242 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
10243 %{
10244 match(Set cr (CmpF src1 src2));
10246 ins_cost(145);
10247 format %{ "ucomiss $src1, $src2\n\t"
10248 "jnp,s exit\n\t"
10249 "pushfq\t# saw NaN, set CF\n\t"
10250 "andq [rsp], #0xffffff2b\n\t"
10251 "popfq\n"
10252 "exit: nop\t# avoid branch to branch" %}
10253 opcode(0x0F, 0x2E);
10254 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10255 cmpfp_fixup);
10256 ins_pipe(pipe_slow);
10257 %}
10259 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
10260 match(Set cr (CmpF src1 src2));
10262 ins_cost(145);
10263 format %{ "ucomiss $src1, $src2" %}
10264 ins_encode %{
10265 __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10266 %}
10267 ins_pipe(pipe_slow);
10268 %}
10270 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
10271 %{
10272 match(Set cr (CmpF src1 (LoadF src2)));
10274 ins_cost(145);
10275 format %{ "ucomiss $src1, $src2\n\t"
10276 "jnp,s exit\n\t"
10277 "pushfq\t# saw NaN, set CF\n\t"
10278 "andq [rsp], #0xffffff2b\n\t"
10279 "popfq\n"
10280 "exit: nop\t# avoid branch to branch" %}
10281 opcode(0x0F, 0x2E);
10282 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10283 cmpfp_fixup);
10284 ins_pipe(pipe_slow);
10285 %}
10287 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
10288 match(Set cr (CmpF src1 (LoadF src2)));
10290 ins_cost(100);
10291 format %{ "ucomiss $src1, $src2" %}
10292 opcode(0x0F, 0x2E);
10293 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2));
10294 ins_pipe(pipe_slow);
10295 %}
10297 instruct cmpF_cc_imm(rFlagsRegU cr, regF src, immF con) %{
10298 match(Set cr (CmpF src con));
10300 ins_cost(145);
10301 format %{ "ucomiss $src, [$constantaddress]\t# load from constant table: float=$con\n\t"
10302 "jnp,s exit\n\t"
10303 "pushfq\t# saw NaN, set CF\n\t"
10304 "andq [rsp], #0xffffff2b\n\t"
10305 "popfq\n"
10306 "exit: nop\t# avoid branch to branch" %}
10307 ins_encode %{
10308 Label L_exit;
10309 __ ucomiss($src$$XMMRegister, $constantaddress($con));
10310 __ jcc(Assembler::noParity, L_exit);
10311 __ pushf();
10312 __ andq(rsp, 0xffffff2b);
10313 __ popf();
10314 __ bind(L_exit);
10315 __ nop();
10316 %}
10317 ins_pipe(pipe_slow);
10318 %}
10320 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src, immF con) %{
10321 match(Set cr (CmpF src con));
10322 ins_cost(100);
10323 format %{ "ucomiss $src, [$constantaddress]\t# load from constant table: float=$con" %}
10324 ins_encode %{
10325 __ ucomiss($src$$XMMRegister, $constantaddress($con));
10326 %}
10327 ins_pipe(pipe_slow);
10328 %}
10330 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
10331 %{
10332 match(Set cr (CmpD src1 src2));
10334 ins_cost(145);
10335 format %{ "ucomisd $src1, $src2\n\t"
10336 "jnp,s exit\n\t"
10337 "pushfq\t# saw NaN, set CF\n\t"
10338 "andq [rsp], #0xffffff2b\n\t"
10339 "popfq\n"
10340 "exit: nop\t# avoid branch to branch" %}
10341 opcode(0x66, 0x0F, 0x2E);
10342 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10343 cmpfp_fixup);
10344 ins_pipe(pipe_slow);
10345 %}
10347 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
10348 match(Set cr (CmpD src1 src2));
10350 ins_cost(100);
10351 format %{ "ucomisd $src1, $src2 test" %}
10352 ins_encode %{
10353 __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
10354 %}
10355 ins_pipe(pipe_slow);
10356 %}
10358 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
10359 %{
10360 match(Set cr (CmpD src1 (LoadD src2)));
10362 ins_cost(145);
10363 format %{ "ucomisd $src1, $src2\n\t"
10364 "jnp,s exit\n\t"
10365 "pushfq\t# saw NaN, set CF\n\t"
10366 "andq [rsp], #0xffffff2b\n\t"
10367 "popfq\n"
10368 "exit: nop\t# avoid branch to branch" %}
10369 opcode(0x66, 0x0F, 0x2E);
10370 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10371 cmpfp_fixup);
10372 ins_pipe(pipe_slow);
10373 %}
10375 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
10376 match(Set cr (CmpD src1 (LoadD src2)));
10378 ins_cost(100);
10379 format %{ "ucomisd $src1, $src2" %}
10380 opcode(0x66, 0x0F, 0x2E);
10381 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2));
10382 ins_pipe(pipe_slow);
10383 %}
10385 instruct cmpD_cc_imm(rFlagsRegU cr, regD src, immD con) %{
10386 match(Set cr (CmpD src con));
10388 ins_cost(145);
10389 format %{ "ucomisd $src, [$constantaddress]\t# load from constant table: double=$con\n\t"
10390 "jnp,s exit\n\t"
10391 "pushfq\t# saw NaN, set CF\n\t"
10392 "andq [rsp], #0xffffff2b\n\t"
10393 "popfq\n"
10394 "exit: nop\t# avoid branch to branch" %}
10395 ins_encode %{
10396 Label L_exit;
10397 __ ucomisd($src$$XMMRegister, $constantaddress($con));
10398 __ jcc(Assembler::noParity, L_exit);
10399 __ pushf();
10400 __ andq(rsp, 0xffffff2b);
10401 __ popf();
10402 __ bind(L_exit);
10403 __ nop();
10404 %}
10405 ins_pipe(pipe_slow);
10406 %}
10408 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src, immD con) %{
10409 match(Set cr (CmpD src con));
10410 ins_cost(100);
10411 format %{ "ucomisd $src, [$constantaddress]\t# load from constant table: double=$con" %}
10412 ins_encode %{
10413 __ ucomisd($src$$XMMRegister, $constantaddress($con));
10414 %}
10415 ins_pipe(pipe_slow);
10416 %}
10418 // Compare into -1,0,1
10419 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
10420 %{
10421 match(Set dst (CmpF3 src1 src2));
10422 effect(KILL cr);
10424 ins_cost(275);
10425 format %{ "ucomiss $src1, $src2\n\t"
10426 "movl $dst, #-1\n\t"
10427 "jp,s done\n\t"
10428 "jb,s done\n\t"
10429 "setne $dst\n\t"
10430 "movzbl $dst, $dst\n"
10431 "done:" %}
10433 opcode(0x0F, 0x2E);
10434 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10435 cmpfp3(dst));
10436 ins_pipe(pipe_slow);
10437 %}
10439 // Compare into -1,0,1
10440 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
10441 %{
10442 match(Set dst (CmpF3 src1 (LoadF src2)));
10443 effect(KILL cr);
10445 ins_cost(275);
10446 format %{ "ucomiss $src1, $src2\n\t"
10447 "movl $dst, #-1\n\t"
10448 "jp,s done\n\t"
10449 "jb,s done\n\t"
10450 "setne $dst\n\t"
10451 "movzbl $dst, $dst\n"
10452 "done:" %}
10454 opcode(0x0F, 0x2E);
10455 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10456 cmpfp3(dst));
10457 ins_pipe(pipe_slow);
10458 %}
10460 // Compare into -1,0,1
10461 instruct cmpF_imm(rRegI dst, regF src, immF con, rFlagsReg cr) %{
10462 match(Set dst (CmpF3 src con));
10463 effect(KILL cr);
10465 ins_cost(275);
10466 format %{ "ucomiss $src, [$constantaddress]\t# load from constant table: float=$con\n\t"
10467 "movl $dst, #-1\n\t"
10468 "jp,s done\n\t"
10469 "jb,s done\n\t"
10470 "setne $dst\n\t"
10471 "movzbl $dst, $dst\n"
10472 "done:" %}
10473 ins_encode %{
10474 Label L_done;
10475 Register Rdst = $dst$$Register;
10476 __ ucomiss($src$$XMMRegister, $constantaddress($con));
10477 __ movl(Rdst, -1);
10478 __ jcc(Assembler::parity, L_done);
10479 __ jcc(Assembler::below, L_done);
10480 __ setb(Assembler::notEqual, Rdst);
10481 __ movzbl(Rdst, Rdst);
10482 __ bind(L_done);
10483 %}
10484 ins_pipe(pipe_slow);
10485 %}
10487 // Compare into -1,0,1
10488 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
10489 %{
10490 match(Set dst (CmpD3 src1 src2));
10491 effect(KILL cr);
10493 ins_cost(275);
10494 format %{ "ucomisd $src1, $src2\n\t"
10495 "movl $dst, #-1\n\t"
10496 "jp,s done\n\t"
10497 "jb,s done\n\t"
10498 "setne $dst\n\t"
10499 "movzbl $dst, $dst\n"
10500 "done:" %}
10502 opcode(0x66, 0x0F, 0x2E);
10503 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10504 cmpfp3(dst));
10505 ins_pipe(pipe_slow);
10506 %}
10508 // Compare into -1,0,1
10509 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
10510 %{
10511 match(Set dst (CmpD3 src1 (LoadD src2)));
10512 effect(KILL cr);
10514 ins_cost(275);
10515 format %{ "ucomisd $src1, $src2\n\t"
10516 "movl $dst, #-1\n\t"
10517 "jp,s done\n\t"
10518 "jb,s done\n\t"
10519 "setne $dst\n\t"
10520 "movzbl $dst, $dst\n"
10521 "done:" %}
10523 opcode(0x66, 0x0F, 0x2E);
10524 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10525 cmpfp3(dst));
10526 ins_pipe(pipe_slow);
10527 %}
10529 // Compare into -1,0,1
10530 instruct cmpD_imm(rRegI dst, regD src, immD con, rFlagsReg cr) %{
10531 match(Set dst (CmpD3 src con));
10532 effect(KILL cr);
10534 ins_cost(275);
10535 format %{ "ucomisd $src, [$constantaddress]\t# load from constant table: double=$con\n\t"
10536 "movl $dst, #-1\n\t"
10537 "jp,s done\n\t"
10538 "jb,s done\n\t"
10539 "setne $dst\n\t"
10540 "movzbl $dst, $dst\n"
10541 "done:" %}
10542 ins_encode %{
10543 Register Rdst = $dst$$Register;
10544 Label L_done;
10545 __ ucomisd($src$$XMMRegister, $constantaddress($con));
10546 __ movl(Rdst, -1);
10547 __ jcc(Assembler::parity, L_done);
10548 __ jcc(Assembler::below, L_done);
10549 __ setb(Assembler::notEqual, Rdst);
10550 __ movzbl(Rdst, Rdst);
10551 __ bind(L_done);
10552 %}
10553 ins_pipe(pipe_slow);
10554 %}
10556 instruct addF_reg(regF dst, regF src)
10557 %{
10558 match(Set dst (AddF dst src));
10560 format %{ "addss $dst, $src" %}
10561 ins_cost(150); // XXX
10562 opcode(0xF3, 0x0F, 0x58);
10563 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10564 ins_pipe(pipe_slow);
10565 %}
10567 instruct addF_mem(regF dst, memory src)
10568 %{
10569 match(Set dst (AddF dst (LoadF src)));
10571 format %{ "addss $dst, $src" %}
10572 ins_cost(150); // XXX
10573 opcode(0xF3, 0x0F, 0x58);
10574 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10575 ins_pipe(pipe_slow);
10576 %}
10578 instruct addF_imm(regF dst, immF con) %{
10579 match(Set dst (AddF dst con));
10580 format %{ "addss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10581 ins_cost(150); // XXX
10582 ins_encode %{
10583 __ addss($dst$$XMMRegister, $constantaddress($con));
10584 %}
10585 ins_pipe(pipe_slow);
10586 %}
10588 instruct addD_reg(regD dst, regD src)
10589 %{
10590 match(Set dst (AddD dst src));
10592 format %{ "addsd $dst, $src" %}
10593 ins_cost(150); // XXX
10594 opcode(0xF2, 0x0F, 0x58);
10595 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10596 ins_pipe(pipe_slow);
10597 %}
10599 instruct addD_mem(regD dst, memory src)
10600 %{
10601 match(Set dst (AddD dst (LoadD src)));
10603 format %{ "addsd $dst, $src" %}
10604 ins_cost(150); // XXX
10605 opcode(0xF2, 0x0F, 0x58);
10606 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10607 ins_pipe(pipe_slow);
10608 %}
10610 instruct addD_imm(regD dst, immD con) %{
10611 match(Set dst (AddD dst con));
10612 format %{ "addsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10613 ins_cost(150); // XXX
10614 ins_encode %{
10615 __ addsd($dst$$XMMRegister, $constantaddress($con));
10616 %}
10617 ins_pipe(pipe_slow);
10618 %}
10620 instruct subF_reg(regF dst, regF src)
10621 %{
10622 match(Set dst (SubF dst src));
10624 format %{ "subss $dst, $src" %}
10625 ins_cost(150); // XXX
10626 opcode(0xF3, 0x0F, 0x5C);
10627 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10628 ins_pipe(pipe_slow);
10629 %}
10631 instruct subF_mem(regF dst, memory src)
10632 %{
10633 match(Set dst (SubF dst (LoadF src)));
10635 format %{ "subss $dst, $src" %}
10636 ins_cost(150); // XXX
10637 opcode(0xF3, 0x0F, 0x5C);
10638 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10639 ins_pipe(pipe_slow);
10640 %}
10642 instruct subF_imm(regF dst, immF con) %{
10643 match(Set dst (SubF dst con));
10644 format %{ "subss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10645 ins_cost(150); // XXX
10646 ins_encode %{
10647 __ subss($dst$$XMMRegister, $constantaddress($con));
10648 %}
10649 ins_pipe(pipe_slow);
10650 %}
10652 instruct subD_reg(regD dst, regD src)
10653 %{
10654 match(Set dst (SubD dst src));
10656 format %{ "subsd $dst, $src" %}
10657 ins_cost(150); // XXX
10658 opcode(0xF2, 0x0F, 0x5C);
10659 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10660 ins_pipe(pipe_slow);
10661 %}
10663 instruct subD_mem(regD dst, memory src)
10664 %{
10665 match(Set dst (SubD dst (LoadD src)));
10667 format %{ "subsd $dst, $src" %}
10668 ins_cost(150); // XXX
10669 opcode(0xF2, 0x0F, 0x5C);
10670 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10671 ins_pipe(pipe_slow);
10672 %}
10674 instruct subD_imm(regD dst, immD con) %{
10675 match(Set dst (SubD dst con));
10676 format %{ "subsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10677 ins_cost(150); // XXX
10678 ins_encode %{
10679 __ subsd($dst$$XMMRegister, $constantaddress($con));
10680 %}
10681 ins_pipe(pipe_slow);
10682 %}
10684 instruct mulF_reg(regF dst, regF src)
10685 %{
10686 match(Set dst (MulF dst src));
10688 format %{ "mulss $dst, $src" %}
10689 ins_cost(150); // XXX
10690 opcode(0xF3, 0x0F, 0x59);
10691 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10692 ins_pipe(pipe_slow);
10693 %}
10695 instruct mulF_mem(regF dst, memory src)
10696 %{
10697 match(Set dst (MulF dst (LoadF src)));
10699 format %{ "mulss $dst, $src" %}
10700 ins_cost(150); // XXX
10701 opcode(0xF3, 0x0F, 0x59);
10702 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10703 ins_pipe(pipe_slow);
10704 %}
10706 instruct mulF_imm(regF dst, immF con) %{
10707 match(Set dst (MulF dst con));
10708 format %{ "mulss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10709 ins_cost(150); // XXX
10710 ins_encode %{
10711 __ mulss($dst$$XMMRegister, $constantaddress($con));
10712 %}
10713 ins_pipe(pipe_slow);
10714 %}
10716 instruct mulD_reg(regD dst, regD src)
10717 %{
10718 match(Set dst (MulD dst src));
10720 format %{ "mulsd $dst, $src" %}
10721 ins_cost(150); // XXX
10722 opcode(0xF2, 0x0F, 0x59);
10723 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10724 ins_pipe(pipe_slow);
10725 %}
10727 instruct mulD_mem(regD dst, memory src)
10728 %{
10729 match(Set dst (MulD dst (LoadD src)));
10731 format %{ "mulsd $dst, $src" %}
10732 ins_cost(150); // XXX
10733 opcode(0xF2, 0x0F, 0x59);
10734 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10735 ins_pipe(pipe_slow);
10736 %}
10738 instruct mulD_imm(regD dst, immD con) %{
10739 match(Set dst (MulD dst con));
10740 format %{ "mulsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10741 ins_cost(150); // XXX
10742 ins_encode %{
10743 __ mulsd($dst$$XMMRegister, $constantaddress($con));
10744 %}
10745 ins_pipe(pipe_slow);
10746 %}
10748 instruct divF_reg(regF dst, regF src)
10749 %{
10750 match(Set dst (DivF dst src));
10752 format %{ "divss $dst, $src" %}
10753 ins_cost(150); // XXX
10754 opcode(0xF3, 0x0F, 0x5E);
10755 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10756 ins_pipe(pipe_slow);
10757 %}
10759 instruct divF_mem(regF dst, memory src)
10760 %{
10761 match(Set dst (DivF dst (LoadF src)));
10763 format %{ "divss $dst, $src" %}
10764 ins_cost(150); // XXX
10765 opcode(0xF3, 0x0F, 0x5E);
10766 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10767 ins_pipe(pipe_slow);
10768 %}
10770 instruct divF_imm(regF dst, immF con) %{
10771 match(Set dst (DivF dst con));
10772 format %{ "divss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10773 ins_cost(150); // XXX
10774 ins_encode %{
10775 __ divss($dst$$XMMRegister, $constantaddress($con));
10776 %}
10777 ins_pipe(pipe_slow);
10778 %}
10780 instruct divD_reg(regD dst, regD src)
10781 %{
10782 match(Set dst (DivD dst src));
10784 format %{ "divsd $dst, $src" %}
10785 ins_cost(150); // XXX
10786 opcode(0xF2, 0x0F, 0x5E);
10787 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10788 ins_pipe(pipe_slow);
10789 %}
10791 instruct divD_mem(regD dst, memory src)
10792 %{
10793 match(Set dst (DivD dst (LoadD src)));
10795 format %{ "divsd $dst, $src" %}
10796 ins_cost(150); // XXX
10797 opcode(0xF2, 0x0F, 0x5E);
10798 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10799 ins_pipe(pipe_slow);
10800 %}
10802 instruct divD_imm(regD dst, immD con) %{
10803 match(Set dst (DivD dst con));
10804 format %{ "divsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10805 ins_cost(150); // XXX
10806 ins_encode %{
10807 __ divsd($dst$$XMMRegister, $constantaddress($con));
10808 %}
10809 ins_pipe(pipe_slow);
10810 %}
10812 instruct sqrtF_reg(regF dst, regF src)
10813 %{
10814 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10816 format %{ "sqrtss $dst, $src" %}
10817 ins_cost(150); // XXX
10818 opcode(0xF3, 0x0F, 0x51);
10819 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10820 ins_pipe(pipe_slow);
10821 %}
10823 instruct sqrtF_mem(regF dst, memory src)
10824 %{
10825 match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
10827 format %{ "sqrtss $dst, $src" %}
10828 ins_cost(150); // XXX
10829 opcode(0xF3, 0x0F, 0x51);
10830 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10831 ins_pipe(pipe_slow);
10832 %}
10834 instruct sqrtF_imm(regF dst, immF con) %{
10835 match(Set dst (ConvD2F (SqrtD (ConvF2D con))));
10836 format %{ "sqrtss $dst, [$constantaddress]\t# load from constant table: float=$con" %}
10837 ins_cost(150); // XXX
10838 ins_encode %{
10839 __ sqrtss($dst$$XMMRegister, $constantaddress($con));
10840 %}
10841 ins_pipe(pipe_slow);
10842 %}
10844 instruct sqrtD_reg(regD dst, regD src)
10845 %{
10846 match(Set dst (SqrtD src));
10848 format %{ "sqrtsd $dst, $src" %}
10849 ins_cost(150); // XXX
10850 opcode(0xF2, 0x0F, 0x51);
10851 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10852 ins_pipe(pipe_slow);
10853 %}
10855 instruct sqrtD_mem(regD dst, memory src)
10856 %{
10857 match(Set dst (SqrtD (LoadD src)));
10859 format %{ "sqrtsd $dst, $src" %}
10860 ins_cost(150); // XXX
10861 opcode(0xF2, 0x0F, 0x51);
10862 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10863 ins_pipe(pipe_slow);
10864 %}
10866 instruct sqrtD_imm(regD dst, immD con) %{
10867 match(Set dst (SqrtD con));
10868 format %{ "sqrtsd $dst, [$constantaddress]\t# load from constant table: double=$con" %}
10869 ins_cost(150); // XXX
10870 ins_encode %{
10871 __ sqrtsd($dst$$XMMRegister, $constantaddress($con));
10872 %}
10873 ins_pipe(pipe_slow);
10874 %}
10876 instruct absF_reg(regF dst)
10877 %{
10878 match(Set dst (AbsF dst));
10880 format %{ "andps $dst, [0x7fffffff]\t# abs float by sign masking" %}
10881 ins_encode(absF_encoding(dst));
10882 ins_pipe(pipe_slow);
10883 %}
10885 instruct absD_reg(regD dst)
10886 %{
10887 match(Set dst (AbsD dst));
10889 format %{ "andpd $dst, [0x7fffffffffffffff]\t"
10890 "# abs double by sign masking" %}
10891 ins_encode(absD_encoding(dst));
10892 ins_pipe(pipe_slow);
10893 %}
10895 instruct negF_reg(regF dst)
10896 %{
10897 match(Set dst (NegF dst));
10899 format %{ "xorps $dst, [0x80000000]\t# neg float by sign flipping" %}
10900 ins_encode(negF_encoding(dst));
10901 ins_pipe(pipe_slow);
10902 %}
10904 instruct negD_reg(regD dst)
10905 %{
10906 match(Set dst (NegD dst));
10908 format %{ "xorpd $dst, [0x8000000000000000]\t"
10909 "# neg double by sign flipping" %}
10910 ins_encode(negD_encoding(dst));
10911 ins_pipe(pipe_slow);
10912 %}
10914 // -----------Trig and Trancendental Instructions------------------------------
10915 instruct cosD_reg(regD dst) %{
10916 match(Set dst (CosD dst));
10918 format %{ "dcos $dst\n\t" %}
10919 opcode(0xD9, 0xFF);
10920 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10921 ins_pipe( pipe_slow );
10922 %}
10924 instruct sinD_reg(regD dst) %{
10925 match(Set dst (SinD dst));
10927 format %{ "dsin $dst\n\t" %}
10928 opcode(0xD9, 0xFE);
10929 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
10930 ins_pipe( pipe_slow );
10931 %}
10933 instruct tanD_reg(regD dst) %{
10934 match(Set dst (TanD dst));
10936 format %{ "dtan $dst\n\t" %}
10937 ins_encode( Push_SrcXD(dst),
10938 Opcode(0xD9), Opcode(0xF2), //fptan
10939 Opcode(0xDD), Opcode(0xD8), //fstp st
10940 Push_ResultXD(dst) );
10941 ins_pipe( pipe_slow );
10942 %}
10944 instruct log10D_reg(regD dst) %{
10945 // The source and result Double operands in XMM registers
10946 match(Set dst (Log10D dst));
10947 // fldlg2 ; push log_10(2) on the FPU stack; full 80-bit number
10948 // fyl2x ; compute log_10(2) * log_2(x)
10949 format %{ "fldlg2\t\t\t#Log10\n\t"
10950 "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
10951 %}
10952 ins_encode(Opcode(0xD9), Opcode(0xEC), // fldlg2
10953 Push_SrcXD(dst),
10954 Opcode(0xD9), Opcode(0xF1), // fyl2x
10955 Push_ResultXD(dst));
10957 ins_pipe( pipe_slow );
10958 %}
10960 instruct logD_reg(regD dst) %{
10961 // The source and result Double operands in XMM registers
10962 match(Set dst (LogD dst));
10963 // fldln2 ; push log_e(2) on the FPU stack; full 80-bit number
10964 // fyl2x ; compute log_e(2) * log_2(x)
10965 format %{ "fldln2\t\t\t#Log_e\n\t"
10966 "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
10967 %}
10968 ins_encode( Opcode(0xD9), Opcode(0xED), // fldln2
10969 Push_SrcXD(dst),
10970 Opcode(0xD9), Opcode(0xF1), // fyl2x
10971 Push_ResultXD(dst));
10972 ins_pipe( pipe_slow );
10973 %}
10977 //----------Arithmetic Conversion Instructions---------------------------------
10979 instruct roundFloat_nop(regF dst)
10980 %{
10981 match(Set dst (RoundFloat dst));
10983 ins_cost(0);
10984 ins_encode();
10985 ins_pipe(empty);
10986 %}
10988 instruct roundDouble_nop(regD dst)
10989 %{
10990 match(Set dst (RoundDouble dst));
10992 ins_cost(0);
10993 ins_encode();
10994 ins_pipe(empty);
10995 %}
10997 instruct convF2D_reg_reg(regD dst, regF src)
10998 %{
10999 match(Set dst (ConvF2D src));
11001 format %{ "cvtss2sd $dst, $src" %}
11002 opcode(0xF3, 0x0F, 0x5A);
11003 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11004 ins_pipe(pipe_slow); // XXX
11005 %}
11007 instruct convF2D_reg_mem(regD dst, memory src)
11008 %{
11009 match(Set dst (ConvF2D (LoadF src)));
11011 format %{ "cvtss2sd $dst, $src" %}
11012 opcode(0xF3, 0x0F, 0x5A);
11013 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11014 ins_pipe(pipe_slow); // XXX
11015 %}
11017 instruct convD2F_reg_reg(regF dst, regD src)
11018 %{
11019 match(Set dst (ConvD2F src));
11021 format %{ "cvtsd2ss $dst, $src" %}
11022 opcode(0xF2, 0x0F, 0x5A);
11023 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11024 ins_pipe(pipe_slow); // XXX
11025 %}
11027 instruct convD2F_reg_mem(regF dst, memory src)
11028 %{
11029 match(Set dst (ConvD2F (LoadD src)));
11031 format %{ "cvtsd2ss $dst, $src" %}
11032 opcode(0xF2, 0x0F, 0x5A);
11033 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11034 ins_pipe(pipe_slow); // XXX
11035 %}
11037 // XXX do mem variants
11038 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
11039 %{
11040 match(Set dst (ConvF2I src));
11041 effect(KILL cr);
11043 format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
11044 "cmpl $dst, #0x80000000\n\t"
11045 "jne,s done\n\t"
11046 "subq rsp, #8\n\t"
11047 "movss [rsp], $src\n\t"
11048 "call f2i_fixup\n\t"
11049 "popq $dst\n"
11050 "done: "%}
11051 opcode(0xF3, 0x0F, 0x2C);
11052 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
11053 f2i_fixup(dst, src));
11054 ins_pipe(pipe_slow);
11055 %}
11057 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
11058 %{
11059 match(Set dst (ConvF2L src));
11060 effect(KILL cr);
11062 format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
11063 "cmpq $dst, [0x8000000000000000]\n\t"
11064 "jne,s done\n\t"
11065 "subq rsp, #8\n\t"
11066 "movss [rsp], $src\n\t"
11067 "call f2l_fixup\n\t"
11068 "popq $dst\n"
11069 "done: "%}
11070 opcode(0xF3, 0x0F, 0x2C);
11071 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
11072 f2l_fixup(dst, src));
11073 ins_pipe(pipe_slow);
11074 %}
11076 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
11077 %{
11078 match(Set dst (ConvD2I src));
11079 effect(KILL cr);
11081 format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
11082 "cmpl $dst, #0x80000000\n\t"
11083 "jne,s done\n\t"
11084 "subq rsp, #8\n\t"
11085 "movsd [rsp], $src\n\t"
11086 "call d2i_fixup\n\t"
11087 "popq $dst\n"
11088 "done: "%}
11089 opcode(0xF2, 0x0F, 0x2C);
11090 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
11091 d2i_fixup(dst, src));
11092 ins_pipe(pipe_slow);
11093 %}
11095 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
11096 %{
11097 match(Set dst (ConvD2L src));
11098 effect(KILL cr);
11100 format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
11101 "cmpq $dst, [0x8000000000000000]\n\t"
11102 "jne,s done\n\t"
11103 "subq rsp, #8\n\t"
11104 "movsd [rsp], $src\n\t"
11105 "call d2l_fixup\n\t"
11106 "popq $dst\n"
11107 "done: "%}
11108 opcode(0xF2, 0x0F, 0x2C);
11109 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
11110 d2l_fixup(dst, src));
11111 ins_pipe(pipe_slow);
11112 %}
11114 instruct convI2F_reg_reg(regF dst, rRegI src)
11115 %{
11116 predicate(!UseXmmI2F);
11117 match(Set dst (ConvI2F src));
11119 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11120 opcode(0xF3, 0x0F, 0x2A);
11121 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11122 ins_pipe(pipe_slow); // XXX
11123 %}
11125 instruct convI2F_reg_mem(regF dst, memory src)
11126 %{
11127 match(Set dst (ConvI2F (LoadI src)));
11129 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11130 opcode(0xF3, 0x0F, 0x2A);
11131 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11132 ins_pipe(pipe_slow); // XXX
11133 %}
11135 instruct convI2D_reg_reg(regD dst, rRegI src)
11136 %{
11137 predicate(!UseXmmI2D);
11138 match(Set dst (ConvI2D src));
11140 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11141 opcode(0xF2, 0x0F, 0x2A);
11142 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11143 ins_pipe(pipe_slow); // XXX
11144 %}
11146 instruct convI2D_reg_mem(regD dst, memory src)
11147 %{
11148 match(Set dst (ConvI2D (LoadI src)));
11150 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11151 opcode(0xF2, 0x0F, 0x2A);
11152 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11153 ins_pipe(pipe_slow); // XXX
11154 %}
11156 instruct convXI2F_reg(regF dst, rRegI src)
11157 %{
11158 predicate(UseXmmI2F);
11159 match(Set dst (ConvI2F src));
11161 format %{ "movdl $dst, $src\n\t"
11162 "cvtdq2psl $dst, $dst\t# i2f" %}
11163 ins_encode %{
11164 __ movdl($dst$$XMMRegister, $src$$Register);
11165 __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
11166 %}
11167 ins_pipe(pipe_slow); // XXX
11168 %}
11170 instruct convXI2D_reg(regD dst, rRegI src)
11171 %{
11172 predicate(UseXmmI2D);
11173 match(Set dst (ConvI2D src));
11175 format %{ "movdl $dst, $src\n\t"
11176 "cvtdq2pdl $dst, $dst\t# i2d" %}
11177 ins_encode %{
11178 __ movdl($dst$$XMMRegister, $src$$Register);
11179 __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
11180 %}
11181 ins_pipe(pipe_slow); // XXX
11182 %}
11184 instruct convL2F_reg_reg(regF dst, rRegL src)
11185 %{
11186 match(Set dst (ConvL2F src));
11188 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11189 opcode(0xF3, 0x0F, 0x2A);
11190 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11191 ins_pipe(pipe_slow); // XXX
11192 %}
11194 instruct convL2F_reg_mem(regF dst, memory src)
11195 %{
11196 match(Set dst (ConvL2F (LoadL src)));
11198 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11199 opcode(0xF3, 0x0F, 0x2A);
11200 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11201 ins_pipe(pipe_slow); // XXX
11202 %}
11204 instruct convL2D_reg_reg(regD dst, rRegL src)
11205 %{
11206 match(Set dst (ConvL2D src));
11208 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11209 opcode(0xF2, 0x0F, 0x2A);
11210 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11211 ins_pipe(pipe_slow); // XXX
11212 %}
11214 instruct convL2D_reg_mem(regD dst, memory src)
11215 %{
11216 match(Set dst (ConvL2D (LoadL src)));
11218 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11219 opcode(0xF2, 0x0F, 0x2A);
11220 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11221 ins_pipe(pipe_slow); // XXX
11222 %}
11224 instruct convI2L_reg_reg(rRegL dst, rRegI src)
11225 %{
11226 match(Set dst (ConvI2L src));
11228 ins_cost(125);
11229 format %{ "movslq $dst, $src\t# i2l" %}
11230 ins_encode %{
11231 __ movslq($dst$$Register, $src$$Register);
11232 %}
11233 ins_pipe(ialu_reg_reg);
11234 %}
11236 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
11237 // %{
11238 // match(Set dst (ConvI2L src));
11239 // // predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
11240 // // _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
11241 // predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
11242 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
11243 // ((const TypeNode*) n)->type()->is_long()->_lo ==
11244 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
11246 // format %{ "movl $dst, $src\t# unsigned i2l" %}
11247 // ins_encode(enc_copy(dst, src));
11248 // // opcode(0x63); // needs REX.W
11249 // // ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
11250 // ins_pipe(ialu_reg_reg);
11251 // %}
11253 // Zero-extend convert int to long
11254 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
11255 %{
11256 match(Set dst (AndL (ConvI2L src) mask));
11258 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11259 ins_encode(enc_copy(dst, src));
11260 ins_pipe(ialu_reg_reg);
11261 %}
11263 // Zero-extend convert int to long
11264 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
11265 %{
11266 match(Set dst (AndL (ConvI2L (LoadI src)) mask));
11268 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11269 opcode(0x8B);
11270 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11271 ins_pipe(ialu_reg_mem);
11272 %}
11274 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
11275 %{
11276 match(Set dst (AndL src mask));
11278 format %{ "movl $dst, $src\t# zero-extend long" %}
11279 ins_encode(enc_copy_always(dst, src));
11280 ins_pipe(ialu_reg_reg);
11281 %}
11283 instruct convL2I_reg_reg(rRegI dst, rRegL src)
11284 %{
11285 match(Set dst (ConvL2I src));
11287 format %{ "movl $dst, $src\t# l2i" %}
11288 ins_encode(enc_copy_always(dst, src));
11289 ins_pipe(ialu_reg_reg);
11290 %}
11293 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
11294 match(Set dst (MoveF2I src));
11295 effect(DEF dst, USE src);
11297 ins_cost(125);
11298 format %{ "movl $dst, $src\t# MoveF2I_stack_reg" %}
11299 opcode(0x8B);
11300 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11301 ins_pipe(ialu_reg_mem);
11302 %}
11304 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
11305 match(Set dst (MoveI2F src));
11306 effect(DEF dst, USE src);
11308 ins_cost(125);
11309 format %{ "movss $dst, $src\t# MoveI2F_stack_reg" %}
11310 opcode(0xF3, 0x0F, 0x10);
11311 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11312 ins_pipe(pipe_slow);
11313 %}
11315 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
11316 match(Set dst (MoveD2L src));
11317 effect(DEF dst, USE src);
11319 ins_cost(125);
11320 format %{ "movq $dst, $src\t# MoveD2L_stack_reg" %}
11321 opcode(0x8B);
11322 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
11323 ins_pipe(ialu_reg_mem);
11324 %}
11326 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
11327 predicate(!UseXmmLoadAndClearUpper);
11328 match(Set dst (MoveL2D src));
11329 effect(DEF dst, USE src);
11331 ins_cost(125);
11332 format %{ "movlpd $dst, $src\t# MoveL2D_stack_reg" %}
11333 opcode(0x66, 0x0F, 0x12);
11334 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11335 ins_pipe(pipe_slow);
11336 %}
11338 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
11339 predicate(UseXmmLoadAndClearUpper);
11340 match(Set dst (MoveL2D src));
11341 effect(DEF dst, USE src);
11343 ins_cost(125);
11344 format %{ "movsd $dst, $src\t# MoveL2D_stack_reg" %}
11345 opcode(0xF2, 0x0F, 0x10);
11346 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11347 ins_pipe(pipe_slow);
11348 %}
11351 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
11352 match(Set dst (MoveF2I src));
11353 effect(DEF dst, USE src);
11355 ins_cost(95); // XXX
11356 format %{ "movss $dst, $src\t# MoveF2I_reg_stack" %}
11357 opcode(0xF3, 0x0F, 0x11);
11358 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11359 ins_pipe(pipe_slow);
11360 %}
11362 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
11363 match(Set dst (MoveI2F src));
11364 effect(DEF dst, USE src);
11366 ins_cost(100);
11367 format %{ "movl $dst, $src\t# MoveI2F_reg_stack" %}
11368 opcode(0x89);
11369 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
11370 ins_pipe( ialu_mem_reg );
11371 %}
11373 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
11374 match(Set dst (MoveD2L src));
11375 effect(DEF dst, USE src);
11377 ins_cost(95); // XXX
11378 format %{ "movsd $dst, $src\t# MoveL2D_reg_stack" %}
11379 opcode(0xF2, 0x0F, 0x11);
11380 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11381 ins_pipe(pipe_slow);
11382 %}
11384 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
11385 match(Set dst (MoveL2D src));
11386 effect(DEF dst, USE src);
11388 ins_cost(100);
11389 format %{ "movq $dst, $src\t# MoveL2D_reg_stack" %}
11390 opcode(0x89);
11391 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
11392 ins_pipe(ialu_mem_reg);
11393 %}
11395 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
11396 match(Set dst (MoveF2I src));
11397 effect(DEF dst, USE src);
11398 ins_cost(85);
11399 format %{ "movd $dst,$src\t# MoveF2I" %}
11400 ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
11401 ins_pipe( pipe_slow );
11402 %}
11404 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
11405 match(Set dst (MoveD2L src));
11406 effect(DEF dst, USE src);
11407 ins_cost(85);
11408 format %{ "movd $dst,$src\t# MoveD2L" %}
11409 ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
11410 ins_pipe( pipe_slow );
11411 %}
11413 // The next instructions have long latency and use Int unit. Set high cost.
11414 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
11415 match(Set dst (MoveI2F src));
11416 effect(DEF dst, USE src);
11417 ins_cost(300);
11418 format %{ "movd $dst,$src\t# MoveI2F" %}
11419 ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
11420 ins_pipe( pipe_slow );
11421 %}
11423 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
11424 match(Set dst (MoveL2D src));
11425 effect(DEF dst, USE src);
11426 ins_cost(300);
11427 format %{ "movd $dst,$src\t# MoveL2D" %}
11428 ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
11429 ins_pipe( pipe_slow );
11430 %}
11432 // Replicate scalar to packed byte (1 byte) values in xmm
11433 instruct Repl8B_reg(regD dst, regD src) %{
11434 match(Set dst (Replicate8B src));
11435 format %{ "MOVDQA $dst,$src\n\t"
11436 "PUNPCKLBW $dst,$dst\n\t"
11437 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11438 ins_encode( pshufd_8x8(dst, src));
11439 ins_pipe( pipe_slow );
11440 %}
11442 // Replicate scalar to packed byte (1 byte) values in xmm
11443 instruct Repl8B_rRegI(regD dst, rRegI src) %{
11444 match(Set dst (Replicate8B src));
11445 format %{ "MOVD $dst,$src\n\t"
11446 "PUNPCKLBW $dst,$dst\n\t"
11447 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11448 ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
11449 ins_pipe( pipe_slow );
11450 %}
11452 // Replicate scalar zero to packed byte (1 byte) values in xmm
11453 instruct Repl8B_immI0(regD dst, immI0 zero) %{
11454 match(Set dst (Replicate8B zero));
11455 format %{ "PXOR $dst,$dst\t! replicate8B" %}
11456 ins_encode( pxor(dst, dst));
11457 ins_pipe( fpu_reg_reg );
11458 %}
11460 // Replicate scalar to packed shore (2 byte) values in xmm
11461 instruct Repl4S_reg(regD dst, regD src) %{
11462 match(Set dst (Replicate4S src));
11463 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
11464 ins_encode( pshufd_4x16(dst, src));
11465 ins_pipe( fpu_reg_reg );
11466 %}
11468 // Replicate scalar to packed shore (2 byte) values in xmm
11469 instruct Repl4S_rRegI(regD dst, rRegI src) %{
11470 match(Set dst (Replicate4S src));
11471 format %{ "MOVD $dst,$src\n\t"
11472 "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
11473 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11474 ins_pipe( fpu_reg_reg );
11475 %}
11477 // Replicate scalar zero to packed short (2 byte) values in xmm
11478 instruct Repl4S_immI0(regD dst, immI0 zero) %{
11479 match(Set dst (Replicate4S zero));
11480 format %{ "PXOR $dst,$dst\t! replicate4S" %}
11481 ins_encode( pxor(dst, dst));
11482 ins_pipe( fpu_reg_reg );
11483 %}
11485 // Replicate scalar to packed char (2 byte) values in xmm
11486 instruct Repl4C_reg(regD dst, regD src) %{
11487 match(Set dst (Replicate4C src));
11488 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
11489 ins_encode( pshufd_4x16(dst, src));
11490 ins_pipe( fpu_reg_reg );
11491 %}
11493 // Replicate scalar to packed char (2 byte) values in xmm
11494 instruct Repl4C_rRegI(regD dst, rRegI src) %{
11495 match(Set dst (Replicate4C src));
11496 format %{ "MOVD $dst,$src\n\t"
11497 "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
11498 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11499 ins_pipe( fpu_reg_reg );
11500 %}
11502 // Replicate scalar zero to packed char (2 byte) values in xmm
11503 instruct Repl4C_immI0(regD dst, immI0 zero) %{
11504 match(Set dst (Replicate4C zero));
11505 format %{ "PXOR $dst,$dst\t! replicate4C" %}
11506 ins_encode( pxor(dst, dst));
11507 ins_pipe( fpu_reg_reg );
11508 %}
11510 // Replicate scalar to packed integer (4 byte) values in xmm
11511 instruct Repl2I_reg(regD dst, regD src) %{
11512 match(Set dst (Replicate2I src));
11513 format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
11514 ins_encode( pshufd(dst, src, 0x00));
11515 ins_pipe( fpu_reg_reg );
11516 %}
11518 // Replicate scalar to packed integer (4 byte) values in xmm
11519 instruct Repl2I_rRegI(regD dst, rRegI src) %{
11520 match(Set dst (Replicate2I src));
11521 format %{ "MOVD $dst,$src\n\t"
11522 "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
11523 ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
11524 ins_pipe( fpu_reg_reg );
11525 %}
11527 // Replicate scalar zero to packed integer (2 byte) values in xmm
11528 instruct Repl2I_immI0(regD dst, immI0 zero) %{
11529 match(Set dst (Replicate2I zero));
11530 format %{ "PXOR $dst,$dst\t! replicate2I" %}
11531 ins_encode( pxor(dst, dst));
11532 ins_pipe( fpu_reg_reg );
11533 %}
11535 // Replicate scalar to packed single precision floating point values in xmm
11536 instruct Repl2F_reg(regD dst, regD src) %{
11537 match(Set dst (Replicate2F src));
11538 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11539 ins_encode( pshufd(dst, src, 0xe0));
11540 ins_pipe( fpu_reg_reg );
11541 %}
11543 // Replicate scalar to packed single precision floating point values in xmm
11544 instruct Repl2F_regF(regD dst, regF src) %{
11545 match(Set dst (Replicate2F src));
11546 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11547 ins_encode( pshufd(dst, src, 0xe0));
11548 ins_pipe( fpu_reg_reg );
11549 %}
11551 // Replicate scalar to packed single precision floating point values in xmm
11552 instruct Repl2F_immF0(regD dst, immF0 zero) %{
11553 match(Set dst (Replicate2F zero));
11554 format %{ "PXOR $dst,$dst\t! replicate2F" %}
11555 ins_encode( pxor(dst, dst));
11556 ins_pipe( fpu_reg_reg );
11557 %}
11560 // =======================================================================
11561 // fast clearing of an array
11562 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
11563 rFlagsReg cr)
11564 %{
11565 match(Set dummy (ClearArray cnt base));
11566 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11568 format %{ "xorl rax, rax\t# ClearArray:\n\t"
11569 "rep stosq\t# Store rax to *rdi++ while rcx--" %}
11570 ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
11571 Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
11572 ins_pipe(pipe_slow);
11573 %}
11575 instruct string_compare(rdi_RegP str1, rcx_RegI cnt1, rsi_RegP str2, rdx_RegI cnt2,
11576 rax_RegI result, regD tmp1, rFlagsReg cr)
11577 %{
11578 match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
11579 effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
11581 format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result // KILL $tmp1" %}
11582 ins_encode %{
11583 __ string_compare($str1$$Register, $str2$$Register,
11584 $cnt1$$Register, $cnt2$$Register, $result$$Register,
11585 $tmp1$$XMMRegister);
11586 %}
11587 ins_pipe( pipe_slow );
11588 %}
11590 // fast search of substring with known size.
11591 instruct string_indexof_con(rdi_RegP str1, rdx_RegI cnt1, rsi_RegP str2, immI int_cnt2,
11592 rbx_RegI result, regD vec, rax_RegI cnt2, rcx_RegI tmp, rFlagsReg cr)
11593 %{
11594 predicate(UseSSE42Intrinsics);
11595 match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
11596 effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, KILL cnt2, KILL tmp, KILL cr);
11598 format %{ "String IndexOf $str1,$cnt1,$str2,$int_cnt2 -> $result // KILL $vec, $cnt1, $cnt2, $tmp" %}
11599 ins_encode %{
11600 int icnt2 = (int)$int_cnt2$$constant;
11601 if (icnt2 >= 8) {
11602 // IndexOf for constant substrings with size >= 8 elements
11603 // which don't need to be loaded through stack.
11604 __ string_indexofC8($str1$$Register, $str2$$Register,
11605 $cnt1$$Register, $cnt2$$Register,
11606 icnt2, $result$$Register,
11607 $vec$$XMMRegister, $tmp$$Register);
11608 } else {
11609 // Small strings are loaded through stack if they cross page boundary.
11610 __ string_indexof($str1$$Register, $str2$$Register,
11611 $cnt1$$Register, $cnt2$$Register,
11612 icnt2, $result$$Register,
11613 $vec$$XMMRegister, $tmp$$Register);
11614 }
11615 %}
11616 ins_pipe( pipe_slow );
11617 %}
11619 instruct string_indexof(rdi_RegP str1, rdx_RegI cnt1, rsi_RegP str2, rax_RegI cnt2,
11620 rbx_RegI result, regD vec, rcx_RegI tmp, rFlagsReg cr)
11621 %{
11622 predicate(UseSSE42Intrinsics);
11623 match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
11624 effect(TEMP vec, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL tmp, KILL cr);
11626 format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result // KILL all" %}
11627 ins_encode %{
11628 __ string_indexof($str1$$Register, $str2$$Register,
11629 $cnt1$$Register, $cnt2$$Register,
11630 (-1), $result$$Register,
11631 $vec$$XMMRegister, $tmp$$Register);
11632 %}
11633 ins_pipe( pipe_slow );
11634 %}
11636 // fast string equals
11637 instruct string_equals(rdi_RegP str1, rsi_RegP str2, rcx_RegI cnt, rax_RegI result,
11638 regD tmp1, regD tmp2, rbx_RegI tmp3, rFlagsReg cr)
11639 %{
11640 match(Set result (StrEquals (Binary str1 str2) cnt));
11641 effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp3, KILL cr);
11643 format %{ "String Equals $str1,$str2,$cnt -> $result // KILL $tmp1, $tmp2, $tmp3" %}
11644 ins_encode %{
11645 __ char_arrays_equals(false, $str1$$Register, $str2$$Register,
11646 $cnt$$Register, $result$$Register, $tmp3$$Register,
11647 $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11648 %}
11649 ins_pipe( pipe_slow );
11650 %}
11652 // fast array equals
11653 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI result,
11654 regD tmp1, regD tmp2, rcx_RegI tmp3, rbx_RegI tmp4, rFlagsReg cr)
11655 %{
11656 match(Set result (AryEq ary1 ary2));
11657 effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
11658 //ins_cost(300);
11660 format %{ "Array Equals $ary1,$ary2 -> $result // KILL $tmp1, $tmp2, $tmp3, $tmp4" %}
11661 ins_encode %{
11662 __ char_arrays_equals(true, $ary1$$Register, $ary2$$Register,
11663 $tmp3$$Register, $result$$Register, $tmp4$$Register,
11664 $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11665 %}
11666 ins_pipe( pipe_slow );
11667 %}
11669 //----------Control Flow Instructions------------------------------------------
11670 // Signed compare Instructions
11672 // XXX more variants!!
11673 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
11674 %{
11675 match(Set cr (CmpI op1 op2));
11676 effect(DEF cr, USE op1, USE op2);
11678 format %{ "cmpl $op1, $op2" %}
11679 opcode(0x3B); /* Opcode 3B /r */
11680 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11681 ins_pipe(ialu_cr_reg_reg);
11682 %}
11684 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
11685 %{
11686 match(Set cr (CmpI op1 op2));
11688 format %{ "cmpl $op1, $op2" %}
11689 opcode(0x81, 0x07); /* Opcode 81 /7 */
11690 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11691 ins_pipe(ialu_cr_reg_imm);
11692 %}
11694 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
11695 %{
11696 match(Set cr (CmpI op1 (LoadI op2)));
11698 ins_cost(500); // XXX
11699 format %{ "cmpl $op1, $op2" %}
11700 opcode(0x3B); /* Opcode 3B /r */
11701 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11702 ins_pipe(ialu_cr_reg_mem);
11703 %}
11705 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
11706 %{
11707 match(Set cr (CmpI src zero));
11709 format %{ "testl $src, $src" %}
11710 opcode(0x85);
11711 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11712 ins_pipe(ialu_cr_reg_imm);
11713 %}
11715 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
11716 %{
11717 match(Set cr (CmpI (AndI src con) zero));
11719 format %{ "testl $src, $con" %}
11720 opcode(0xF7, 0x00);
11721 ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
11722 ins_pipe(ialu_cr_reg_imm);
11723 %}
11725 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
11726 %{
11727 match(Set cr (CmpI (AndI src (LoadI mem)) zero));
11729 format %{ "testl $src, $mem" %}
11730 opcode(0x85);
11731 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
11732 ins_pipe(ialu_cr_reg_mem);
11733 %}
11735 // Unsigned compare Instructions; really, same as signed except they
11736 // produce an rFlagsRegU instead of rFlagsReg.
11737 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
11738 %{
11739 match(Set cr (CmpU op1 op2));
11741 format %{ "cmpl $op1, $op2\t# unsigned" %}
11742 opcode(0x3B); /* Opcode 3B /r */
11743 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11744 ins_pipe(ialu_cr_reg_reg);
11745 %}
11747 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
11748 %{
11749 match(Set cr (CmpU op1 op2));
11751 format %{ "cmpl $op1, $op2\t# unsigned" %}
11752 opcode(0x81,0x07); /* Opcode 81 /7 */
11753 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11754 ins_pipe(ialu_cr_reg_imm);
11755 %}
11757 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
11758 %{
11759 match(Set cr (CmpU op1 (LoadI op2)));
11761 ins_cost(500); // XXX
11762 format %{ "cmpl $op1, $op2\t# unsigned" %}
11763 opcode(0x3B); /* Opcode 3B /r */
11764 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11765 ins_pipe(ialu_cr_reg_mem);
11766 %}
11768 // // // Cisc-spilled version of cmpU_rReg
11769 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
11770 // //%{
11771 // // match(Set cr (CmpU (LoadI op1) op2));
11772 // //
11773 // // format %{ "CMPu $op1,$op2" %}
11774 // // ins_cost(500);
11775 // // opcode(0x39); /* Opcode 39 /r */
11776 // // ins_encode( OpcP, reg_mem( op1, op2) );
11777 // //%}
11779 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
11780 %{
11781 match(Set cr (CmpU src zero));
11783 format %{ "testl $src, $src\t# unsigned" %}
11784 opcode(0x85);
11785 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11786 ins_pipe(ialu_cr_reg_imm);
11787 %}
11789 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
11790 %{
11791 match(Set cr (CmpP op1 op2));
11793 format %{ "cmpq $op1, $op2\t# ptr" %}
11794 opcode(0x3B); /* Opcode 3B /r */
11795 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11796 ins_pipe(ialu_cr_reg_reg);
11797 %}
11799 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
11800 %{
11801 match(Set cr (CmpP op1 (LoadP op2)));
11803 ins_cost(500); // XXX
11804 format %{ "cmpq $op1, $op2\t# ptr" %}
11805 opcode(0x3B); /* Opcode 3B /r */
11806 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11807 ins_pipe(ialu_cr_reg_mem);
11808 %}
11810 // // // Cisc-spilled version of cmpP_rReg
11811 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
11812 // //%{
11813 // // match(Set cr (CmpP (LoadP op1) op2));
11814 // //
11815 // // format %{ "CMPu $op1,$op2" %}
11816 // // ins_cost(500);
11817 // // opcode(0x39); /* Opcode 39 /r */
11818 // // ins_encode( OpcP, reg_mem( op1, op2) );
11819 // //%}
11821 // XXX this is generalized by compP_rReg_mem???
11822 // Compare raw pointer (used in out-of-heap check).
11823 // Only works because non-oop pointers must be raw pointers
11824 // and raw pointers have no anti-dependencies.
11825 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
11826 %{
11827 predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
11828 match(Set cr (CmpP op1 (LoadP op2)));
11830 format %{ "cmpq $op1, $op2\t# raw ptr" %}
11831 opcode(0x3B); /* Opcode 3B /r */
11832 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11833 ins_pipe(ialu_cr_reg_mem);
11834 %}
11836 // This will generate a signed flags result. This should be OK since
11837 // any compare to a zero should be eq/neq.
11838 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
11839 %{
11840 match(Set cr (CmpP src zero));
11842 format %{ "testq $src, $src\t# ptr" %}
11843 opcode(0x85);
11844 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11845 ins_pipe(ialu_cr_reg_imm);
11846 %}
11848 // This will generate a signed flags result. This should be OK since
11849 // any compare to a zero should be eq/neq.
11850 instruct testP_mem(rFlagsReg cr, memory op, immP0 zero)
11851 %{
11852 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
11853 match(Set cr (CmpP (LoadP op) zero));
11855 ins_cost(500); // XXX
11856 format %{ "testq $op, 0xffffffffffffffff\t# ptr" %}
11857 opcode(0xF7); /* Opcode F7 /0 */
11858 ins_encode(REX_mem_wide(op),
11859 OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
11860 ins_pipe(ialu_cr_reg_imm);
11861 %}
11863 instruct testP_mem_reg0(rFlagsReg cr, memory mem, immP0 zero)
11864 %{
11865 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
11866 match(Set cr (CmpP (LoadP mem) zero));
11868 format %{ "cmpq R12, $mem\t# ptr (R12_heapbase==0)" %}
11869 ins_encode %{
11870 __ cmpq(r12, $mem$$Address);
11871 %}
11872 ins_pipe(ialu_cr_reg_mem);
11873 %}
11875 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2)
11876 %{
11877 match(Set cr (CmpN op1 op2));
11879 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11880 ins_encode %{ __ cmpl($op1$$Register, $op2$$Register); %}
11881 ins_pipe(ialu_cr_reg_reg);
11882 %}
11884 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem)
11885 %{
11886 match(Set cr (CmpN src (LoadN mem)));
11888 format %{ "cmpl $src, $mem\t# compressed ptr" %}
11889 ins_encode %{
11890 __ cmpl($src$$Register, $mem$$Address);
11891 %}
11892 ins_pipe(ialu_cr_reg_mem);
11893 %}
11895 instruct compN_rReg_imm(rFlagsRegU cr, rRegN op1, immN op2) %{
11896 match(Set cr (CmpN op1 op2));
11898 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11899 ins_encode %{
11900 __ cmp_narrow_oop($op1$$Register, (jobject)$op2$$constant);
11901 %}
11902 ins_pipe(ialu_cr_reg_imm);
11903 %}
11905 instruct compN_mem_imm(rFlagsRegU cr, memory mem, immN src)
11906 %{
11907 match(Set cr (CmpN src (LoadN mem)));
11909 format %{ "cmpl $mem, $src\t# compressed ptr" %}
11910 ins_encode %{
11911 __ cmp_narrow_oop($mem$$Address, (jobject)$src$$constant);
11912 %}
11913 ins_pipe(ialu_cr_reg_mem);
11914 %}
11916 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
11917 match(Set cr (CmpN src zero));
11919 format %{ "testl $src, $src\t# compressed ptr" %}
11920 ins_encode %{ __ testl($src$$Register, $src$$Register); %}
11921 ins_pipe(ialu_cr_reg_imm);
11922 %}
11924 instruct testN_mem(rFlagsReg cr, memory mem, immN0 zero)
11925 %{
11926 predicate(Universe::narrow_oop_base() != NULL);
11927 match(Set cr (CmpN (LoadN mem) zero));
11929 ins_cost(500); // XXX
11930 format %{ "testl $mem, 0xffffffff\t# compressed ptr" %}
11931 ins_encode %{
11932 __ cmpl($mem$$Address, (int)0xFFFFFFFF);
11933 %}
11934 ins_pipe(ialu_cr_reg_mem);
11935 %}
11937 instruct testN_mem_reg0(rFlagsReg cr, memory mem, immN0 zero)
11938 %{
11939 predicate(Universe::narrow_oop_base() == NULL);
11940 match(Set cr (CmpN (LoadN mem) zero));
11942 format %{ "cmpl R12, $mem\t# compressed ptr (R12_heapbase==0)" %}
11943 ins_encode %{
11944 __ cmpl(r12, $mem$$Address);
11945 %}
11946 ins_pipe(ialu_cr_reg_mem);
11947 %}
11949 // Yanked all unsigned pointer compare operations.
11950 // Pointer compares are done with CmpP which is already unsigned.
11952 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
11953 %{
11954 match(Set cr (CmpL op1 op2));
11956 format %{ "cmpq $op1, $op2" %}
11957 opcode(0x3B); /* Opcode 3B /r */
11958 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11959 ins_pipe(ialu_cr_reg_reg);
11960 %}
11962 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
11963 %{
11964 match(Set cr (CmpL op1 op2));
11966 format %{ "cmpq $op1, $op2" %}
11967 opcode(0x81, 0x07); /* Opcode 81 /7 */
11968 ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
11969 ins_pipe(ialu_cr_reg_imm);
11970 %}
11972 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
11973 %{
11974 match(Set cr (CmpL op1 (LoadL op2)));
11976 format %{ "cmpq $op1, $op2" %}
11977 opcode(0x3B); /* Opcode 3B /r */
11978 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11979 ins_pipe(ialu_cr_reg_mem);
11980 %}
11982 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
11983 %{
11984 match(Set cr (CmpL src zero));
11986 format %{ "testq $src, $src" %}
11987 opcode(0x85);
11988 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11989 ins_pipe(ialu_cr_reg_imm);
11990 %}
11992 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
11993 %{
11994 match(Set cr (CmpL (AndL src con) zero));
11996 format %{ "testq $src, $con\t# long" %}
11997 opcode(0xF7, 0x00);
11998 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
11999 ins_pipe(ialu_cr_reg_imm);
12000 %}
12002 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
12003 %{
12004 match(Set cr (CmpL (AndL src (LoadL mem)) zero));
12006 format %{ "testq $src, $mem" %}
12007 opcode(0x85);
12008 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
12009 ins_pipe(ialu_cr_reg_mem);
12010 %}
12012 // Manifest a CmpL result in an integer register. Very painful.
12013 // This is the test to avoid.
12014 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
12015 %{
12016 match(Set dst (CmpL3 src1 src2));
12017 effect(KILL flags);
12019 ins_cost(275); // XXX
12020 format %{ "cmpq $src1, $src2\t# CmpL3\n\t"
12021 "movl $dst, -1\n\t"
12022 "jl,s done\n\t"
12023 "setne $dst\n\t"
12024 "movzbl $dst, $dst\n\t"
12025 "done:" %}
12026 ins_encode(cmpl3_flag(src1, src2, dst));
12027 ins_pipe(pipe_slow);
12028 %}
12030 //----------Max and Min--------------------------------------------------------
12031 // Min Instructions
12033 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
12034 %{
12035 effect(USE_DEF dst, USE src, USE cr);
12037 format %{ "cmovlgt $dst, $src\t# min" %}
12038 opcode(0x0F, 0x4F);
12039 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
12040 ins_pipe(pipe_cmov_reg);
12041 %}
12044 instruct minI_rReg(rRegI dst, rRegI src)
12045 %{
12046 match(Set dst (MinI dst src));
12048 ins_cost(200);
12049 expand %{
12050 rFlagsReg cr;
12051 compI_rReg(cr, dst, src);
12052 cmovI_reg_g(dst, src, cr);
12053 %}
12054 %}
12056 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
12057 %{
12058 effect(USE_DEF dst, USE src, USE cr);
12060 format %{ "cmovllt $dst, $src\t# max" %}
12061 opcode(0x0F, 0x4C);
12062 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
12063 ins_pipe(pipe_cmov_reg);
12064 %}
12067 instruct maxI_rReg(rRegI dst, rRegI src)
12068 %{
12069 match(Set dst (MaxI dst src));
12071 ins_cost(200);
12072 expand %{
12073 rFlagsReg cr;
12074 compI_rReg(cr, dst, src);
12075 cmovI_reg_l(dst, src, cr);
12076 %}
12077 %}
12079 // ============================================================================
12080 // Branch Instructions
12082 // Jump Direct - Label defines a relative address from JMP+1
12083 instruct jmpDir(label labl)
12084 %{
12085 match(Goto);
12086 effect(USE labl);
12088 ins_cost(300);
12089 format %{ "jmp $labl" %}
12090 size(5);
12091 opcode(0xE9);
12092 ins_encode(OpcP, Lbl(labl));
12093 ins_pipe(pipe_jmp);
12094 ins_pc_relative(1);
12095 %}
12097 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12098 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
12099 %{
12100 match(If cop cr);
12101 effect(USE labl);
12103 ins_cost(300);
12104 format %{ "j$cop $labl" %}
12105 size(6);
12106 opcode(0x0F, 0x80);
12107 ins_encode(Jcc(cop, labl));
12108 ins_pipe(pipe_jcc);
12109 ins_pc_relative(1);
12110 %}
12112 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12113 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
12114 %{
12115 match(CountedLoopEnd cop cr);
12116 effect(USE labl);
12118 ins_cost(300);
12119 format %{ "j$cop $labl\t# loop end" %}
12120 size(6);
12121 opcode(0x0F, 0x80);
12122 ins_encode(Jcc(cop, labl));
12123 ins_pipe(pipe_jcc);
12124 ins_pc_relative(1);
12125 %}
12127 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12128 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12129 match(CountedLoopEnd cop cmp);
12130 effect(USE labl);
12132 ins_cost(300);
12133 format %{ "j$cop,u $labl\t# loop end" %}
12134 size(6);
12135 opcode(0x0F, 0x80);
12136 ins_encode(Jcc(cop, labl));
12137 ins_pipe(pipe_jcc);
12138 ins_pc_relative(1);
12139 %}
12141 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12142 match(CountedLoopEnd cop cmp);
12143 effect(USE labl);
12145 ins_cost(200);
12146 format %{ "j$cop,u $labl\t# loop end" %}
12147 size(6);
12148 opcode(0x0F, 0x80);
12149 ins_encode(Jcc(cop, labl));
12150 ins_pipe(pipe_jcc);
12151 ins_pc_relative(1);
12152 %}
12154 // Jump Direct Conditional - using unsigned comparison
12155 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12156 match(If cop cmp);
12157 effect(USE labl);
12159 ins_cost(300);
12160 format %{ "j$cop,u $labl" %}
12161 size(6);
12162 opcode(0x0F, 0x80);
12163 ins_encode(Jcc(cop, labl));
12164 ins_pipe(pipe_jcc);
12165 ins_pc_relative(1);
12166 %}
12168 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12169 match(If cop cmp);
12170 effect(USE labl);
12172 ins_cost(200);
12173 format %{ "j$cop,u $labl" %}
12174 size(6);
12175 opcode(0x0F, 0x80);
12176 ins_encode(Jcc(cop, labl));
12177 ins_pipe(pipe_jcc);
12178 ins_pc_relative(1);
12179 %}
12181 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12182 match(If cop cmp);
12183 effect(USE labl);
12185 ins_cost(200);
12186 format %{ $$template
12187 if ($cop$$cmpcode == Assembler::notEqual) {
12188 $$emit$$"jp,u $labl\n\t"
12189 $$emit$$"j$cop,u $labl"
12190 } else {
12191 $$emit$$"jp,u done\n\t"
12192 $$emit$$"j$cop,u $labl\n\t"
12193 $$emit$$"done:"
12194 }
12195 %}
12196 size(12);
12197 opcode(0x0F, 0x80);
12198 ins_encode %{
12199 Label* l = $labl$$label;
12200 $$$emit8$primary;
12201 emit_cc(cbuf, $secondary, Assembler::parity);
12202 int parity_disp = -1;
12203 if ($cop$$cmpcode == Assembler::notEqual) {
12204 // the two jumps 6 bytes apart so the jump distances are too
12205 parity_disp = l ? (l->loc_pos() - (cbuf.insts_size() + 4)) : 0;
12206 } else if ($cop$$cmpcode == Assembler::equal) {
12207 parity_disp = 6;
12208 } else {
12209 ShouldNotReachHere();
12210 }
12211 emit_d32(cbuf, parity_disp);
12212 $$$emit8$primary;
12213 emit_cc(cbuf, $secondary, $cop$$cmpcode);
12214 int disp = l ? (l->loc_pos() - (cbuf.insts_size() + 4)) : 0;
12215 emit_d32(cbuf, disp);
12216 %}
12217 ins_pipe(pipe_jcc);
12218 ins_pc_relative(1);
12219 %}
12221 // ============================================================================
12222 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary
12223 // superklass array for an instance of the superklass. Set a hidden
12224 // internal cache on a hit (cache is checked with exposed code in
12225 // gen_subtype_check()). Return NZ for a miss or zero for a hit. The
12226 // encoding ALSO sets flags.
12228 instruct partialSubtypeCheck(rdi_RegP result,
12229 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12230 rFlagsReg cr)
12231 %{
12232 match(Set result (PartialSubtypeCheck sub super));
12233 effect(KILL rcx, KILL cr);
12235 ins_cost(1100); // slightly larger than the next version
12236 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12237 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12238 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12239 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
12240 "jne,s miss\t\t# Missed: rdi not-zero\n\t"
12241 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12242 "xorq $result, $result\t\t Hit: rdi zero\n\t"
12243 "miss:\t" %}
12245 opcode(0x1); // Force a XOR of RDI
12246 ins_encode(enc_PartialSubtypeCheck());
12247 ins_pipe(pipe_slow);
12248 %}
12250 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
12251 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12252 immP0 zero,
12253 rdi_RegP result)
12254 %{
12255 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
12256 effect(KILL rcx, KILL result);
12258 ins_cost(1000);
12259 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12260 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12261 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12262 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
12263 "jne,s miss\t\t# Missed: flags nz\n\t"
12264 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12265 "miss:\t" %}
12267 opcode(0x0); // No need to XOR RDI
12268 ins_encode(enc_PartialSubtypeCheck());
12269 ins_pipe(pipe_slow);
12270 %}
12272 // ============================================================================
12273 // Branch Instructions -- short offset versions
12274 //
12275 // These instructions are used to replace jumps of a long offset (the default
12276 // match) with jumps of a shorter offset. These instructions are all tagged
12277 // with the ins_short_branch attribute, which causes the ADLC to suppress the
12278 // match rules in general matching. Instead, the ADLC generates a conversion
12279 // method in the MachNode which can be used to do in-place replacement of the
12280 // long variant with the shorter variant. The compiler will determine if a
12281 // branch can be taken by the is_short_branch_offset() predicate in the machine
12282 // specific code section of the file.
12284 // Jump Direct - Label defines a relative address from JMP+1
12285 instruct jmpDir_short(label labl) %{
12286 match(Goto);
12287 effect(USE labl);
12289 ins_cost(300);
12290 format %{ "jmp,s $labl" %}
12291 size(2);
12292 opcode(0xEB);
12293 ins_encode(OpcP, LblShort(labl));
12294 ins_pipe(pipe_jmp);
12295 ins_pc_relative(1);
12296 ins_short_branch(1);
12297 %}
12299 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12300 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{
12301 match(If cop cr);
12302 effect(USE labl);
12304 ins_cost(300);
12305 format %{ "j$cop,s $labl" %}
12306 size(2);
12307 opcode(0x70);
12308 ins_encode(JccShort(cop, labl));
12309 ins_pipe(pipe_jcc);
12310 ins_pc_relative(1);
12311 ins_short_branch(1);
12312 %}
12314 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12315 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{
12316 match(CountedLoopEnd cop cr);
12317 effect(USE labl);
12319 ins_cost(300);
12320 format %{ "j$cop,s $labl\t# loop end" %}
12321 size(2);
12322 opcode(0x70);
12323 ins_encode(JccShort(cop, labl));
12324 ins_pipe(pipe_jcc);
12325 ins_pc_relative(1);
12326 ins_short_branch(1);
12327 %}
12329 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12330 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12331 match(CountedLoopEnd cop cmp);
12332 effect(USE labl);
12334 ins_cost(300);
12335 format %{ "j$cop,us $labl\t# loop end" %}
12336 size(2);
12337 opcode(0x70);
12338 ins_encode(JccShort(cop, labl));
12339 ins_pipe(pipe_jcc);
12340 ins_pc_relative(1);
12341 ins_short_branch(1);
12342 %}
12344 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12345 match(CountedLoopEnd cop cmp);
12346 effect(USE labl);
12348 ins_cost(300);
12349 format %{ "j$cop,us $labl\t# loop end" %}
12350 size(2);
12351 opcode(0x70);
12352 ins_encode(JccShort(cop, labl));
12353 ins_pipe(pipe_jcc);
12354 ins_pc_relative(1);
12355 ins_short_branch(1);
12356 %}
12358 // Jump Direct Conditional - using unsigned comparison
12359 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12360 match(If cop cmp);
12361 effect(USE labl);
12363 ins_cost(300);
12364 format %{ "j$cop,us $labl" %}
12365 size(2);
12366 opcode(0x70);
12367 ins_encode(JccShort(cop, labl));
12368 ins_pipe(pipe_jcc);
12369 ins_pc_relative(1);
12370 ins_short_branch(1);
12371 %}
12373 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12374 match(If cop cmp);
12375 effect(USE labl);
12377 ins_cost(300);
12378 format %{ "j$cop,us $labl" %}
12379 size(2);
12380 opcode(0x70);
12381 ins_encode(JccShort(cop, labl));
12382 ins_pipe(pipe_jcc);
12383 ins_pc_relative(1);
12384 ins_short_branch(1);
12385 %}
12387 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12388 match(If cop cmp);
12389 effect(USE labl);
12391 ins_cost(300);
12392 format %{ $$template
12393 if ($cop$$cmpcode == Assembler::notEqual) {
12394 $$emit$$"jp,u,s $labl\n\t"
12395 $$emit$$"j$cop,u,s $labl"
12396 } else {
12397 $$emit$$"jp,u,s done\n\t"
12398 $$emit$$"j$cop,u,s $labl\n\t"
12399 $$emit$$"done:"
12400 }
12401 %}
12402 size(4);
12403 opcode(0x70);
12404 ins_encode %{
12405 Label* l = $labl$$label;
12406 emit_cc(cbuf, $primary, Assembler::parity);
12407 int parity_disp = -1;
12408 if ($cop$$cmpcode == Assembler::notEqual) {
12409 parity_disp = l ? (l->loc_pos() - (cbuf.insts_size() + 1)) : 0;
12410 } else if ($cop$$cmpcode == Assembler::equal) {
12411 parity_disp = 2;
12412 } else {
12413 ShouldNotReachHere();
12414 }
12415 emit_d8(cbuf, parity_disp);
12416 emit_cc(cbuf, $primary, $cop$$cmpcode);
12417 int disp = l ? (l->loc_pos() - (cbuf.insts_size() + 1)) : 0;
12418 emit_d8(cbuf, disp);
12419 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
12420 assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp");
12421 %}
12422 ins_pipe(pipe_jcc);
12423 ins_pc_relative(1);
12424 ins_short_branch(1);
12425 %}
12427 // ============================================================================
12428 // inlined locking and unlocking
12430 instruct cmpFastLock(rFlagsReg cr,
12431 rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
12432 %{
12433 match(Set cr (FastLock object box));
12434 effect(TEMP tmp, TEMP scr);
12436 ins_cost(300);
12437 format %{ "fastlock $object,$box,$tmp,$scr" %}
12438 ins_encode(Fast_Lock(object, box, tmp, scr));
12439 ins_pipe(pipe_slow);
12440 ins_pc_relative(1);
12441 %}
12443 instruct cmpFastUnlock(rFlagsReg cr,
12444 rRegP object, rax_RegP box, rRegP tmp)
12445 %{
12446 match(Set cr (FastUnlock object box));
12447 effect(TEMP tmp);
12449 ins_cost(300);
12450 format %{ "fastunlock $object, $box, $tmp" %}
12451 ins_encode(Fast_Unlock(object, box, tmp));
12452 ins_pipe(pipe_slow);
12453 ins_pc_relative(1);
12454 %}
12457 // ============================================================================
12458 // Safepoint Instructions
12459 instruct safePoint_poll(rFlagsReg cr)
12460 %{
12461 predicate(!Assembler::is_polling_page_far());
12462 match(SafePoint);
12463 effect(KILL cr);
12465 format %{ "testl rax, [rip + #offset_to_poll_page]\t"
12466 "# Safepoint: poll for GC" %}
12467 ins_cost(125);
12468 ins_encode %{
12469 AddressLiteral addr(os::get_polling_page(), relocInfo::poll_type);
12470 __ testl(rax, addr);
12471 %}
12472 ins_pipe(ialu_reg_mem);
12473 %}
12475 instruct safePoint_poll_far(rFlagsReg cr, rRegP poll)
12476 %{
12477 predicate(Assembler::is_polling_page_far());
12478 match(SafePoint poll);
12479 effect(KILL cr, USE poll);
12481 format %{ "testl rax, [$poll]\t"
12482 "# Safepoint: poll for GC" %}
12483 ins_cost(125);
12484 ins_encode %{
12485 __ relocate(relocInfo::poll_type);
12486 __ testl(rax, Address($poll$$Register, 0));
12487 %}
12488 ins_pipe(ialu_reg_mem);
12489 %}
12491 // ============================================================================
12492 // Procedure Call/Return Instructions
12493 // Call Java Static Instruction
12494 // Note: If this code changes, the corresponding ret_addr_offset() and
12495 // compute_padding() functions will have to be adjusted.
12496 instruct CallStaticJavaDirect(method meth) %{
12497 match(CallStaticJava);
12498 predicate(!((CallStaticJavaNode*) n)->is_method_handle_invoke());
12499 effect(USE meth);
12501 ins_cost(300);
12502 format %{ "call,static " %}
12503 opcode(0xE8); /* E8 cd */
12504 ins_encode(Java_Static_Call(meth), call_epilog);
12505 ins_pipe(pipe_slow);
12506 ins_pc_relative(1);
12507 ins_alignment(4);
12508 %}
12510 // Call Java Static Instruction (method handle version)
12511 // Note: If this code changes, the corresponding ret_addr_offset() and
12512 // compute_padding() functions will have to be adjusted.
12513 instruct CallStaticJavaHandle(method meth, rbp_RegP rbp_mh_SP_save) %{
12514 match(CallStaticJava);
12515 predicate(((CallStaticJavaNode*) n)->is_method_handle_invoke());
12516 effect(USE meth);
12517 // RBP is saved by all callees (for interpreter stack correction).
12518 // We use it here for a similar purpose, in {preserve,restore}_SP.
12520 ins_cost(300);
12521 format %{ "call,static/MethodHandle " %}
12522 opcode(0xE8); /* E8 cd */
12523 ins_encode(preserve_SP,
12524 Java_Static_Call(meth),
12525 restore_SP,
12526 call_epilog);
12527 ins_pipe(pipe_slow);
12528 ins_pc_relative(1);
12529 ins_alignment(4);
12530 %}
12532 // Call Java Dynamic Instruction
12533 // Note: If this code changes, the corresponding ret_addr_offset() and
12534 // compute_padding() functions will have to be adjusted.
12535 instruct CallDynamicJavaDirect(method meth)
12536 %{
12537 match(CallDynamicJava);
12538 effect(USE meth);
12540 ins_cost(300);
12541 format %{ "movq rax, #Universe::non_oop_word()\n\t"
12542 "call,dynamic " %}
12543 opcode(0xE8); /* E8 cd */
12544 ins_encode(Java_Dynamic_Call(meth), call_epilog);
12545 ins_pipe(pipe_slow);
12546 ins_pc_relative(1);
12547 ins_alignment(4);
12548 %}
12550 // Call Runtime Instruction
12551 instruct CallRuntimeDirect(method meth)
12552 %{
12553 match(CallRuntime);
12554 effect(USE meth);
12556 ins_cost(300);
12557 format %{ "call,runtime " %}
12558 opcode(0xE8); /* E8 cd */
12559 ins_encode(Java_To_Runtime(meth));
12560 ins_pipe(pipe_slow);
12561 ins_pc_relative(1);
12562 %}
12564 // Call runtime without safepoint
12565 instruct CallLeafDirect(method meth)
12566 %{
12567 match(CallLeaf);
12568 effect(USE meth);
12570 ins_cost(300);
12571 format %{ "call_leaf,runtime " %}
12572 opcode(0xE8); /* E8 cd */
12573 ins_encode(Java_To_Runtime(meth));
12574 ins_pipe(pipe_slow);
12575 ins_pc_relative(1);
12576 %}
12578 // Call runtime without safepoint
12579 instruct CallLeafNoFPDirect(method meth)
12580 %{
12581 match(CallLeafNoFP);
12582 effect(USE meth);
12584 ins_cost(300);
12585 format %{ "call_leaf_nofp,runtime " %}
12586 opcode(0xE8); /* E8 cd */
12587 ins_encode(Java_To_Runtime(meth));
12588 ins_pipe(pipe_slow);
12589 ins_pc_relative(1);
12590 %}
12592 // Return Instruction
12593 // Remove the return address & jump to it.
12594 // Notice: We always emit a nop after a ret to make sure there is room
12595 // for safepoint patching
12596 instruct Ret()
12597 %{
12598 match(Return);
12600 format %{ "ret" %}
12601 opcode(0xC3);
12602 ins_encode(OpcP);
12603 ins_pipe(pipe_jmp);
12604 %}
12606 // Tail Call; Jump from runtime stub to Java code.
12607 // Also known as an 'interprocedural jump'.
12608 // Target of jump will eventually return to caller.
12609 // TailJump below removes the return address.
12610 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
12611 %{
12612 match(TailCall jump_target method_oop);
12614 ins_cost(300);
12615 format %{ "jmp $jump_target\t# rbx holds method oop" %}
12616 opcode(0xFF, 0x4); /* Opcode FF /4 */
12617 ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
12618 ins_pipe(pipe_jmp);
12619 %}
12621 // Tail Jump; remove the return address; jump to target.
12622 // TailCall above leaves the return address around.
12623 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
12624 %{
12625 match(TailJump jump_target ex_oop);
12627 ins_cost(300);
12628 format %{ "popq rdx\t# pop return address\n\t"
12629 "jmp $jump_target" %}
12630 opcode(0xFF, 0x4); /* Opcode FF /4 */
12631 ins_encode(Opcode(0x5a), // popq rdx
12632 REX_reg(jump_target), OpcP, reg_opc(jump_target));
12633 ins_pipe(pipe_jmp);
12634 %}
12636 // Create exception oop: created by stack-crawling runtime code.
12637 // Created exception is now available to this handler, and is setup
12638 // just prior to jumping to this handler. No code emitted.
12639 instruct CreateException(rax_RegP ex_oop)
12640 %{
12641 match(Set ex_oop (CreateEx));
12643 size(0);
12644 // use the following format syntax
12645 format %{ "# exception oop is in rax; no code emitted" %}
12646 ins_encode();
12647 ins_pipe(empty);
12648 %}
12650 // Rethrow exception:
12651 // The exception oop will come in the first argument position.
12652 // Then JUMP (not call) to the rethrow stub code.
12653 instruct RethrowException()
12654 %{
12655 match(Rethrow);
12657 // use the following format syntax
12658 format %{ "jmp rethrow_stub" %}
12659 ins_encode(enc_rethrow);
12660 ins_pipe(pipe_jmp);
12661 %}
12664 //----------PEEPHOLE RULES-----------------------------------------------------
12665 // These must follow all instruction definitions as they use the names
12666 // defined in the instructions definitions.
12667 //
12668 // peepmatch ( root_instr_name [preceding_instruction]* );
12669 //
12670 // peepconstraint %{
12671 // (instruction_number.operand_name relational_op instruction_number.operand_name
12672 // [, ...] );
12673 // // instruction numbers are zero-based using left to right order in peepmatch
12674 //
12675 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
12676 // // provide an instruction_number.operand_name for each operand that appears
12677 // // in the replacement instruction's match rule
12678 //
12679 // ---------VM FLAGS---------------------------------------------------------
12680 //
12681 // All peephole optimizations can be turned off using -XX:-OptoPeephole
12682 //
12683 // Each peephole rule is given an identifying number starting with zero and
12684 // increasing by one in the order seen by the parser. An individual peephole
12685 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
12686 // on the command-line.
12687 //
12688 // ---------CURRENT LIMITATIONS----------------------------------------------
12689 //
12690 // Only match adjacent instructions in same basic block
12691 // Only equality constraints
12692 // Only constraints between operands, not (0.dest_reg == RAX_enc)
12693 // Only one replacement instruction
12694 //
12695 // ---------EXAMPLE----------------------------------------------------------
12696 //
12697 // // pertinent parts of existing instructions in architecture description
12698 // instruct movI(rRegI dst, rRegI src)
12699 // %{
12700 // match(Set dst (CopyI src));
12701 // %}
12702 //
12703 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
12704 // %{
12705 // match(Set dst (AddI dst src));
12706 // effect(KILL cr);
12707 // %}
12708 //
12709 // // Change (inc mov) to lea
12710 // peephole %{
12711 // // increment preceeded by register-register move
12712 // peepmatch ( incI_rReg movI );
12713 // // require that the destination register of the increment
12714 // // match the destination register of the move
12715 // peepconstraint ( 0.dst == 1.dst );
12716 // // construct a replacement instruction that sets
12717 // // the destination to ( move's source register + one )
12718 // peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
12719 // %}
12720 //
12722 // Implementation no longer uses movX instructions since
12723 // machine-independent system no longer uses CopyX nodes.
12724 //
12725 // peephole
12726 // %{
12727 // peepmatch (incI_rReg movI);
12728 // peepconstraint (0.dst == 1.dst);
12729 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12730 // %}
12732 // peephole
12733 // %{
12734 // peepmatch (decI_rReg movI);
12735 // peepconstraint (0.dst == 1.dst);
12736 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12737 // %}
12739 // peephole
12740 // %{
12741 // peepmatch (addI_rReg_imm movI);
12742 // peepconstraint (0.dst == 1.dst);
12743 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12744 // %}
12746 // peephole
12747 // %{
12748 // peepmatch (incL_rReg movL);
12749 // peepconstraint (0.dst == 1.dst);
12750 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12751 // %}
12753 // peephole
12754 // %{
12755 // peepmatch (decL_rReg movL);
12756 // peepconstraint (0.dst == 1.dst);
12757 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12758 // %}
12760 // peephole
12761 // %{
12762 // peepmatch (addL_rReg_imm movL);
12763 // peepconstraint (0.dst == 1.dst);
12764 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12765 // %}
12767 // peephole
12768 // %{
12769 // peepmatch (addP_rReg_imm movP);
12770 // peepconstraint (0.dst == 1.dst);
12771 // peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
12772 // %}
12774 // // Change load of spilled value to only a spill
12775 // instruct storeI(memory mem, rRegI src)
12776 // %{
12777 // match(Set mem (StoreI mem src));
12778 // %}
12779 //
12780 // instruct loadI(rRegI dst, memory mem)
12781 // %{
12782 // match(Set dst (LoadI mem));
12783 // %}
12784 //
12786 peephole
12787 %{
12788 peepmatch (loadI storeI);
12789 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12790 peepreplace (storeI(1.mem 1.mem 1.src));
12791 %}
12793 peephole
12794 %{
12795 peepmatch (loadL storeL);
12796 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12797 peepreplace (storeL(1.mem 1.mem 1.src));
12798 %}
12800 //----------SMARTSPILL RULES---------------------------------------------------
12801 // These must follow all instruction definitions as they use the names
12802 // defined in the instructions definitions.